@yanlinglabs/winter-provider-catalog 0.0.23 → 0.0.27

package/dist/index.js

@@ -48,6 +48,12 @@ var CONTINUATIONS = ["none", "plaintext", "opaque-provider-state", "server-respo
 var READABLE_STATES = ["none", "summary", "full-exposed"];
 var REPLAY_SCOPES = ["current-tool-loop", "current-turn", "selected-turns", "all-turns"];
 var TOOL_LOOP_REQUIREMENTS = ["hard-error", "silent-degradation", "not-required"];
+var EFFORT_REQUEST_FIELDS = ["output_config.effort"];
+var BLOCK_BINDING_BETAS = ["thinking-binding-controls-2026-08-01"];
+var PER_MESSAGE_EFFORT_BETAS = ["mid-conversation-output-config-2026-07-01"];
+var PER_MESSAGE_EFFORT_ITEMS = ["configuration_update"];
+var MID_CONVERSATION_TOOL_CHANGE_BETAS = ["mid-conversation-tool-changes-2026-07-01"];
+var INLINE_TOOL_DEFINITION_BETAS = ["inline-tools-2026-09-15"];
 var CATALOG_VOCABULARIES = {
   protocols: PROTOCOLS,
   authKinds: AUTH_KINDS,
@@ -71,7 +77,13 @@ var CATALOG_VOCABULARIES = {
   continuations: CONTINUATIONS,
   readableStates: READABLE_STATES,
   replayScopes: REPLAY_SCOPES,
-  toolLoopRequirements: TOOL_LOOP_REQUIREMENTS
+  toolLoopRequirements: TOOL_LOOP_REQUIREMENTS,
+  effortRequestFields: EFFORT_REQUEST_FIELDS,
+  blockBindingBetas: BLOCK_BINDING_BETAS,
+  perMessageEffortBetas: PER_MESSAGE_EFFORT_BETAS,
+  perMessageEffortItems: PER_MESSAGE_EFFORT_ITEMS,
+  midConversationToolChangeBetas: MID_CONVERSATION_TOOL_CHANGE_BETAS,
+  inlineToolDefinitionBetas: INLINE_TOOL_DEFINITION_BETAS
 };
 var SECRET_FIELD_NAME_RE = /^(?:api[_-]?key|apikey|secret|secret[_-]?key|password|passwd|token|access[_-]?token|refresh[_-]?token|id[_-]?token|bearer|private[_-]?key|client[_-]?secret|session[_-]?token|credential|credentials|authorization|auth[_-]?token|aws[_-]?secret[_-]?access[_-]?key|aws[_-]?access[_-]?key[_-]?id)$/i;
 var SECRET_VALUE_PATTERNS = [
@@ -286,6 +298,35 @@ function checkReasoning(errs, v, path) {
     if (typeof val !== "string" || !TOOL_LOOP_REQUIREMENTS.includes(val))
       errs.add(p, `unknown tool-loop requirement ${describe(val)}`);
   }, false);
+  checkEvidence(errs, v["effortRequest"], `${path}.effortRequest`, (val, p) => {
+    if (!isRecord(val))
+      return errs.add(p, `expected {field}, got ${describe(val)}`);
+    for (const key of Object.keys(val))
+      if (key !== "field")
+        errs.add(`${p}.${key}`, "unknown key");
+    if (typeof val["field"] !== "string" || !EFFORT_REQUEST_FIELDS.includes(val["field"]))
+      errs.add(`${p}.field`, `unknown effort request field ${describe(val["field"])}`);
+  }, false);
+  checkEvidence(errs, v["blockBinding"], `${path}.blockBinding`, (val, p) => {
+    if (!isRecord(val))
+      return errs.add(p, `expected {beta}, got ${describe(val)}`);
+    for (const key of Object.keys(val))
+      if (key !== "beta")
+        errs.add(`${p}.${key}`, "unknown key");
+    if (typeof val["beta"] !== "string" || !BLOCK_BINDING_BETAS.includes(val["beta"]))
+      errs.add(`${p}.beta`, `unknown block-binding beta ${describe(val["beta"])}`);
+  }, false);
+  checkEvidence(errs, v["perMessageEffort"], `${path}.perMessageEffort`, (val, p) => {
+    if (!isRecord(val))
+      return errs.add(p, `expected {beta} or {item}, got ${describe(val)}`);
+    const keys = Object.keys(val);
+    if (keys.length !== 1 || keys[0] !== "beta" && keys[0] !== "item")
+      return errs.add(p, `expected exactly one of {beta} or {item}, got keys ${describe(keys)}`);
+    if (keys[0] === "beta" && (typeof val["beta"] !== "string" || !PER_MESSAGE_EFFORT_BETAS.includes(val["beta"])))
+      errs.add(`${p}.beta`, `unknown per-message effort beta ${describe(val["beta"])}`);
+    if (keys[0] === "item" && (typeof val["item"] !== "string" || !PER_MESSAGE_EFFORT_ITEMS.includes(val["item"])))
+      errs.add(`${p}.item`, `unknown per-message effort item ${describe(val["item"])}`);
+  }, false);
 }
 function checkProvider(errs, v, path) {
   if (!isRecord(v)) {
@@ -526,6 +567,24 @@ function checkModel(errs, v, path) {
   checkEvidence(errs, v["parallelTools"], `${path}.parallelTools`, evidenceBoolean, false);
   checkEvidence(errs, v["structuredOutput"], `${path}.structuredOutput`, evidenceBoolean, false);
   checkEvidence(errs, v["promptCaching"], `${path}.promptCaching`, evidenceBoolean, false);
+  checkEvidence(errs, v["deferredToolLoading"], `${path}.deferredToolLoading`, evidenceBoolean, false);
+  checkEvidence(errs, v["midConversationSystem"], `${path}.midConversationSystem`, evidenceBoolean, false);
+  checkEvidence(errs, v["promptCacheKey"], `${path}.promptCacheKey`, evidenceBoolean, false);
+  const betaOnly = (vocabulary, what) => (val, p) => {
+    if (!isRecord(val))
+      return errs.add(p, `expected {beta}, got ${describe(val)}`);
+    for (const key of Object.keys(val))
+      if (key !== "beta")
+        errs.add(`${p}.${key}`, "unknown key");
+    if (typeof val["beta"] !== "string" || !vocabulary.includes(val["beta"]))
+      errs.add(`${p}.beta`, `unknown ${what} beta ${describe(val["beta"])}`);
+  };
+  checkEvidence(errs, v["midConversationToolChanges"], `${path}.midConversationToolChanges`, betaOnly(MID_CONVERSATION_TOOL_CHANGE_BETAS, "mid-conversation tool-change"), false);
+  checkEvidence(errs, v["inlineToolDefinitions"], `${path}.inlineToolDefinitions`, betaOnly(INLINE_TOOL_DEFINITION_BETAS, "inline tool-definition"), false);
+  checkEvidence(errs, v["clientToolSearch"], `${path}.clientToolSearch`, evidenceBoolean, false);
+  checkEvidence(errs, v["additionalToolsItem"], `${path}.additionalToolsItem`, evidenceBoolean, false);
+  checkEvidence(errs, v["allowedToolsChoice"], `${path}.allowedToolsChoice`, evidenceBoolean, false);
+  checkEvidence(errs, v["assistantPrefill"], `${path}.assistantPrefill`, evidenceBoolean, false);
   checkEvidence(errs, v["classifierEligible"], `${path}.classifierEligible`, evidenceBoolean, false);
   checkEvidence(errs, v["pricing"], `${path}.pricing`, (val, p) => checkPricing(errs, val, p), false);
   if (v["reasoning"] !== undefined)
@@ -7754,6 +7813,7 @@ var catalog_default = {
       id: "xai",
       displayName: "xAI (Grok)",
       protocols: [
+        "openai-responses",
         "openai-chat-completions"
       ],
       authKinds: [
@@ -7764,7 +7824,7 @@ var catalog_default = {
       },
       modelDiscovery: "openai-models",
       liveCatalogAuthority: "unknown",
-      adapterId: "winter.openai-chat-completions",
+      adapterId: "winter.openai-responses",
       family: "openai",
       upstream: {
         project: "winter",
@@ -10709,13 +10769,13 @@ var catalog_default = {
       },
       pricing: {
         value: {
-          inputPerMTokUsd: 0.144,
-          outputPerMTokUsd: 0.574
+          inputPerMTokUsd: 0.149,
+          outputPerMTokUsd: 0.596
         },
         source: "official-doc",
-        sourceRef: "https://www.alibabacloud.com/help/en/model-studio/qwen3-coder-next — China (Beijing) qwen3-coder-next: input $0.144, output $0.574 per million tokens; ≤32K input; >32K–128K $0.216/$0.861; >128K–256K $0.359/$1.434",
-        confidence: "declared",
-        observedAt: "2026-09-25T09:40:00Z"
+        sourceRef: "https://help.aliyun.com/zh/model-studio/model-pricing — Beijing CNY list, qwen3-coder-next, ≤32K; 32K–128K: ¥1.5/¥6; 128K–256K: ¥2.5/¥10 input tier: ¥1 input/¥4 output per million; USD = CNY × 0.1490 (CNY→USD reference rate for 2026-09-25, https://www.investing.com/currencies/cny-usd-historical-data) — the one rate every CNY-derived price in this catalog uses; not an Alibaba FX quote.",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
       },
       unsupportedParameters: [],
       status: "candidate",
@@ -10817,13 +10877,13 @@ var catalog_default = {
       },
       pricing: {
         value: {
-          inputPerMTokUsd: 0.574,
-          outputPerMTokUsd: 2.294
+          inputPerMTokUsd: 0.596,
+          outputPerMTokUsd: 2.384
         },
         source: "official-doc",
-        sourceRef: "https://www.alibabacloud.com/help/en/model-studio/qwen3-coder-plus — China (Beijing) qwen3-coder-plus: input $0.574, output $2.294 per million tokens; ≤32K input; >32K–128K $0.861/$3.441; >128K–256K $1.434/$5.735; >256K–1M $2.868/$28.671",
-        confidence: "declared",
-        observedAt: "2026-09-25T09:40:00Z"
+        sourceRef: "https://help.aliyun.com/zh/model-studio/model-pricing — Beijing CNY list, qwen3-coder-plus, ≤32K; 32K–128K: ¥6/¥24; 128K–256K: ¥10/¥40; 256K–1M: ¥20/¥200 input tier: ¥4 input/¥16 output per million; USD = CNY × 0.1490 (CNY→USD reference rate for 2026-09-25, https://www.investing.com/currencies/cny-usd-historical-data) — the one rate every CNY-derived price in this catalog uses; not an Alibaba FX quote.",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
       },
       unsupportedParameters: [],
       status: "candidate",
@@ -10927,13 +10987,13 @@ var catalog_default = {
       },
       pricing: {
         value: {
-          inputPerMTokUsd: 0.115,
-          outputPerMTokUsd: 0.917
+          inputPerMTokUsd: 0.1192,
+          outputPerMTokUsd: 0.9536
         },
         source: "official-doc",
-        sourceRef: "https://www.alibabacloud.com/help/en/model-studio/qwen3-5-122b-a10b — China (Beijing) qwen3.5-122b-a10b: input $0.115, output $0.917 per million tokens; ≤128K input; >128K–256K $0.287/$2.294",
-        confidence: "declared",
-        observedAt: "2026-09-25T09:40:00Z"
+        sourceRef: "https://help.aliyun.com/zh/model-studio/model-pricing — Beijing CNY list, qwen3.5-122b-a10b, ≤128K; >128K–256K: ¥2/¥16 input tier: ¥0.8 input/¥6.4 output per million; USD = CNY × 0.1490 (CNY→USD reference rate for 2026-09-25, https://www.investing.com/currencies/cny-usd-historical-data) — the one rate every CNY-derived price in this catalog uses; not an Alibaba FX quote.",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
       },
       unsupportedParameters: [],
       status: "candidate",
@@ -11037,13 +11097,13 @@ var catalog_default = {
       },
       pricing: {
         value: {
-          inputPerMTokUsd: 0.172,
-          outputPerMTokUsd: 1.032
+          inputPerMTokUsd: 0.1788,
+          outputPerMTokUsd: 1.0728
         },
         source: "official-doc",
-        sourceRef: "https://www.alibabacloud.com/help/en/model-studio/qwen3-5-397b-a17b — China (Beijing) qwen3.5-397b-a17b: input $0.172, output $1.032 per million tokens; ≤128K input; >128K–256K $0.43/$2.58",
-        confidence: "declared",
-        observedAt: "2026-09-25T09:40:00Z"
+        sourceRef: "https://help.aliyun.com/zh/model-studio/model-pricing — Beijing CNY list, qwen3.5-397b-a17b, ≤128K; >128K–256K: ¥3/¥18 input tier: ¥1.2 input/¥7.2 output per million; USD = CNY × 0.1490 (CNY→USD reference rate for 2026-09-25, https://www.investing.com/currencies/cny-usd-historical-data) — the one rate every CNY-derived price in this catalog uses; not an Alibaba FX quote.",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
       },
       unsupportedParameters: [],
       status: "candidate",
@@ -11133,13 +11193,13 @@ var catalog_default = {
       },
       pricing: {
         value: {
-          inputPerMTokUsd: 0.115,
-          outputPerMTokUsd: 0.688
+          inputPerMTokUsd: 0.1192,
+          outputPerMTokUsd: 0.7152
         },
         source: "official-doc",
-        sourceRef: "https://www.alibabacloud.com/help/en/model-studio/model-pricing — China (Beijing) qwen3.5-plus: input $0.115, output $0.688 per million tokens; ≤128K input; >128K–256K $0.287/$1.72; >256K–1M $0.573/$3.44",
-        confidence: "declared",
-        observedAt: "2026-09-25T09:40:00Z"
+        sourceRef: "https://help.aliyun.com/zh/model-studio/model-pricing — Beijing CNY list, qwen3.5-plus, ≤128K; 128K–256K: ¥2/¥12; 256K–1M: ¥4/¥24 input tier: ¥0.8 input/¥4.8 output per million; USD = CNY × 0.1490 (CNY→USD reference rate for 2026-09-25, https://www.investing.com/currencies/cny-usd-historical-data) — the one rate every CNY-derived price in this catalog uses; not an Alibaba FX quote.",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
       },
       unsupportedParameters: [],
       status: "candidate",
@@ -11243,13 +11303,13 @@ var catalog_default = {
       },
       pricing: {
         value: {
-          inputPerMTokUsd: 0.412564,
-          outputPerMTokUsd: 2.475384
+          inputPerMTokUsd: 0.447,
+          outputPerMTokUsd: 2.682
         },
         source: "official-doc",
-        sourceRef: "https://www.alibabacloud.com/help/en/model-studio/qwen3-6-27b — China (Beijing) qwen3.6-27b: input $0.412564, output $2.475384 per million tokens; all input",
-        confidence: "declared",
-        observedAt: "2026-09-25T09:40:00Z"
+        sourceRef: "https://help.aliyun.com/zh/model-studio/model-pricing — Beijing CNY list, qwen3.6-27b, ≤256K input tier: ¥3 input/¥18 output per million; USD = CNY × 0.1490 (CNY→USD reference rate for 2026-09-25, https://www.investing.com/currencies/cny-usd-historical-data) — the one rate every CNY-derived price in this catalog uses; not an Alibaba FX quote.",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
       },
       unsupportedParameters: [],
       status: "candidate",
@@ -11346,13 +11406,13 @@ var catalog_default = {
       },
       pricing: {
         value: {
-          inputPerMTokUsd: 0.165,
-          outputPerMTokUsd: 0.99
+          inputPerMTokUsd: 0.1788,
+          outputPerMTokUsd: 1.0728
         },
         source: "official-doc",
-        sourceRef: "https://www.alibabacloud.com/help/en/model-studio/model-pricing — China (Beijing) qwen3.6-flash: input $0.165, output $0.99 per million tokens; ≤256K input; >256K–1M $0.66/$3.961",
-        confidence: "declared",
-        observedAt: "2026-09-25T09:40:00Z"
+        sourceRef: "https://help.aliyun.com/zh/model-studio/model-pricing — Beijing CNY list, qwen3.6-flash, ≤256K; 256K–1M: ¥4.8/¥28.8 input tier: ¥1.2 input/¥7.2 output per million; USD = CNY × 0.1490 (CNY→USD reference rate for 2026-09-25, https://www.investing.com/currencies/cny-usd-historical-data) — the one rate every CNY-derived price in this catalog uses; not an Alibaba FX quote.",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
       },
       unsupportedParameters: [],
       status: "candidate",
@@ -11442,13 +11502,13 @@ var catalog_default = {
       },
       pricing: {
         value: {
-          inputPerMTokUsd: 0.276,
-          outputPerMTokUsd: 1.651
+          inputPerMTokUsd: 0.298,
+          outputPerMTokUsd: 1.788
         },
         source: "official-doc",
-        sourceRef: "https://www.alibabacloud.com/help/en/model-studio/model-pricing — China (Beijing) qwen3.6-plus: input $0.276, output $1.651 per million tokens; ≤256K input; >256K–1M $1.101/$6.602",
-        confidence: "declared",
-        observedAt: "2026-09-25T09:40:00Z"
+        sourceRef: "https://help.aliyun.com/zh/model-studio/model-pricing — Beijing CNY list, qwen3.6-plus, ≤256K; >256K–1M: ¥8/¥48 input tier: ¥2 input/¥12 output per million; USD = CNY × 0.1490 (CNY→USD reference rate for 2026-09-25, https://www.investing.com/currencies/cny-usd-historical-data) — the one rate every CNY-derived price in this catalog uses; not an Alibaba FX quote.",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
       },
       unsupportedParameters: [],
       status: "candidate",
@@ -11536,13 +11596,13 @@ var catalog_default = {
       },
       pricing: {
         value: {
-          inputPerMTokUsd: 1.65,
-          outputPerMTokUsd: 4.951
+          inputPerMTokUsd: 1.788,
+          outputPerMTokUsd: 5.364
         },
         source: "official-doc",
-        sourceRef: "https://www.alibabacloud.com/help/en/model-studio/model-pricing — China (Beijing) qwen3.7-max: input $1.65, output $4.951 per million tokens; 0–1M input tokens",
-        confidence: "declared",
-        observedAt: "2026-09-25T09:40:00Z"
+        sourceRef: "https://help.aliyun.com/zh/model-studio/model-pricing — Beijing CNY list, qwen3.7-max, 0–1M input tier: ¥12 input/¥36 output per million; USD = CNY × 0.1490 (CNY→USD reference rate for 2026-09-25, https://www.investing.com/currencies/cny-usd-historical-data) — the one rate every CNY-derived price in this catalog uses; not an Alibaba FX quote.",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
       },
       unsupportedParameters: [],
       status: "candidate",
@@ -11639,13 +11699,13 @@ var catalog_default = {
       },
       pricing: {
         value: {
-          inputPerMTokUsd: 0.276,
-          outputPerMTokUsd: 1.101
+          inputPerMTokUsd: 0.298,
+          outputPerMTokUsd: 1.192
         },
         source: "official-doc",
-        sourceRef: "https://www.alibabacloud.com/help/en/model-studio/model-pricing — China (Beijing) qwen3.7-plus: input $0.276, output $1.101 per million tokens; ≤256K input; >256K–1M $0.826/$3.301; list price before limited-time 20% discount",
-        confidence: "declared",
-        observedAt: "2026-09-25T09:40:00Z"
+        sourceRef: "https://help.aliyun.com/zh/model-studio/model-pricing — Beijing CNY list, qwen3.7-plus, ≤256K; >256K–1M: ¥6/¥24; base list price before temporary 20% promotion input tier: ¥2 input/¥8 output per million; USD = CNY × 0.1490 (CNY→USD reference rate for 2026-09-25, https://www.investing.com/currencies/cny-usd-historical-data) — the one rate every CNY-derived price in this catalog uses; not an Alibaba FX quote.",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
       },
       unsupportedParameters: [],
       status: "candidate",
@@ -11735,13 +11795,13 @@ var catalog_default = {
       },
       pricing: {
         value: {
-          inputPerMTokUsd: 0.113,
-          outputPerMTokUsd: 0.382
+          inputPerMTokUsd: 0.1192,
+          outputPerMTokUsd: 0.4023
         },
         source: "official-doc",
-        sourceRef: "https://www.alibabacloud.com/help/en/model-studio/model-pricing — China (Beijing) qwen3.8-flash: input $0.113, output $0.382 per million tokens; 0–1M input tokens",
-        confidence: "declared",
-        observedAt: "2026-09-25T09:40:00Z"
+        sourceRef: "https://help.aliyun.com/zh/model-studio/model-pricing — Beijing CNY list, qwen3.8-flash, 0–1M input tier: ¥0.8 input/¥2.7 output per million; USD = CNY × 0.1490 (CNY→USD reference rate for 2026-09-25, https://www.investing.com/currencies/cny-usd-historical-data) — the one rate every CNY-derived price in this catalog uses; not an Alibaba FX quote.",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
       },
       unsupportedParameters: [],
       status: "candidate",
@@ -11831,13 +11891,13 @@ var catalog_default = {
       },
       pricing: {
         value: {
-          inputPerMTokUsd: 1.65,
-          outputPerMTokUsd: 4.951
+          inputPerMTokUsd: 1.788,
+          outputPerMTokUsd: 5.364
         },
         source: "official-doc",
-        sourceRef: "https://www.alibabacloud.com/help/en/model-studio/model-pricing — China (Beijing) qwen3.8-max: input $1.65, output $4.951 per million tokens; 0–1M input tokens",
-        confidence: "declared",
-        observedAt: "2026-09-25T09:40:00Z"
+        sourceRef: "https://help.aliyun.com/zh/model-studio/model-pricing — Beijing CNY list, qwen3.8-max, 0–1M input tier: ¥12 input/¥36 output per million; USD = CNY × 0.1490 (CNY→USD reference rate for 2026-09-25, https://www.investing.com/currencies/cny-usd-historical-data) — the one rate every CNY-derived price in this catalog uses; not an Alibaba FX quote.",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
       },
       unsupportedParameters: [],
       status: "candidate",
@@ -12163,13 +12223,13 @@ var catalog_default = {
       },
       pricing: {
         value: {
-          inputPerMTokUsd: 0.144,
-          outputPerMTokUsd: 0.574
+          inputPerMTokUsd: 0.149,
+          outputPerMTokUsd: 0.596
         },
         source: "official-doc",
-        sourceRef: "https://www.alibabacloud.com/help/en/model-studio/qwen3-coder-next — China (Beijing) qwen3-coder-next: input $0.144, output $0.574 per million tokens; ≤32K input; >32K–128K $0.216/$0.861; >128K–256K $0.359/$1.434",
-        confidence: "declared",
-        observedAt: "2026-09-25T09:40:00Z"
+        sourceRef: "https://help.aliyun.com/zh/model-studio/model-pricing — Beijing CNY list, qwen3-coder-next, ≤32K; 32K–128K: ¥1.5/¥6; 128K–256K: ¥2.5/¥10 input tier: ¥1 input/¥4 output per million; USD = CNY × 0.1490 (CNY→USD reference rate for 2026-09-25, https://www.investing.com/currencies/cny-usd-historical-data) — the one rate every CNY-derived price in this catalog uses; not an Alibaba FX quote.",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
       },
       unsupportedParameters: [],
       status: "candidate",
@@ -12265,13 +12325,13 @@ var catalog_default = {
       },
       pricing: {
         value: {
-          inputPerMTokUsd: 0.574,
-          outputPerMTokUsd: 2.294
+          inputPerMTokUsd: 0.596,
+          outputPerMTokUsd: 2.384
         },
         source: "official-doc",
-        sourceRef: "https://www.alibabacloud.com/help/en/model-studio/qwen3-coder-plus — China (Beijing) qwen3-coder-plus: input $0.574, output $2.294 per million tokens; ≤32K input; >32K–128K $0.861/$3.441; >128K–256K $1.434/$5.735; >256K–1M $2.868/$28.671",
-        confidence: "declared",
-        observedAt: "2026-09-25T09:40:00Z"
+        sourceRef: "https://help.aliyun.com/zh/model-studio/model-pricing — Beijing CNY list, qwen3-coder-plus, ≤32K; 32K–128K: ¥6/¥24; 128K–256K: ¥10/¥40; 256K–1M: ¥20/¥200 input tier: ¥4 input/¥16 output per million; USD = CNY × 0.1490 (CNY→USD reference rate for 2026-09-25, https://www.investing.com/currencies/cny-usd-historical-data) — the one rate every CNY-derived price in this catalog uses; not an Alibaba FX quote.",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
       },
       unsupportedParameters: [],
       status: "candidate",
@@ -12868,13 +12928,13 @@ var catalog_default = {
       },
       pricing: {
         value: {
-          inputPerMTokUsd: 0.165,
-          outputPerMTokUsd: 0.99
+          inputPerMTokUsd: 0.1788,
+          outputPerMTokUsd: 1.0728
         },
         source: "official-doc",
-        sourceRef: "https://www.alibabacloud.com/help/en/model-studio/model-pricing — China (Beijing) qwen3.6-flash: input $0.165, output $0.99 per million tokens; ≤256K input; >256K–1M $0.66/$3.961",
-        confidence: "declared",
-        observedAt: "2026-09-25T09:40:00Z"
+        sourceRef: "https://help.aliyun.com/zh/model-studio/model-pricing — Beijing CNY list, qwen3.6-flash, ≤256K; 256K–1M: ¥4.8/¥28.8 input tier: ¥1.2 input/¥7.2 output per million; USD = CNY × 0.1490 (CNY→USD reference rate for 2026-09-25, https://www.investing.com/currencies/cny-usd-historical-data) — the one rate every CNY-derived price in this catalog uses; not an Alibaba FX quote.",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
       },
       unsupportedParameters: [],
       status: "candidate",
@@ -13239,13 +13299,13 @@ var catalog_default = {
       },
       pricing: {
         value: {
-          inputPerMTokUsd: 0.113,
-          outputPerMTokUsd: 0.382
+          inputPerMTokUsd: 0.1192,
+          outputPerMTokUsd: 0.4023
         },
         source: "official-doc",
-        sourceRef: "https://www.alibabacloud.com/help/en/model-studio/model-pricing — China (Beijing) qwen3.8-flash: input $0.113, output $0.382 per million tokens; 0–1M input tokens",
-        confidence: "declared",
-        observedAt: "2026-09-25T09:40:00Z"
+        sourceRef: "https://help.aliyun.com/zh/model-studio/model-pricing — Beijing CNY list, qwen3.8-flash, 0–1M input tier: ¥0.8 input/¥2.7 output per million; USD = CNY × 0.1490 (CNY→USD reference rate for 2026-09-25, https://www.investing.com/currencies/cny-usd-historical-data) — the one rate every CNY-derived price in this catalog uses; not an Alibaba FX quote.",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
       },
       unsupportedParameters: [],
       status: "candidate",
@@ -15083,7 +15143,16 @@ var catalog_default = {
           "max"
         ],
         continuation: "opaque-provider-state",
-        defaultEffort: "high"
+        defaultEffort: "high",
+        effortRequest: {
+          value: {
+            field: "output_config.effort"
+          },
+          source: "official-doc",
+          confidence: "declared",
+          observedAt: "2026-09-25T12:30:00Z",
+          sourceRef: "https://platform.claude.com/docs/en/build-with-claude/effort — the effort page's supportedModels list names claude-fable-5-1, claude-fable-5, claude-opus-5-5, claude-opus-5, claude-opus-4-8, claude-opus-4-7, claude-opus-4-6, claude-opus-4-5-20251101, claude-sonnet-5 and claude-sonnet-4-6: effort is requested through `output_config.effort` (Claude 4.7 and later reject a manual `thinking.budget_tokens`, https://platform.claude.com/docs/en/build-with-claude/thinking-troubleshooting#rejected-configurations)."
+        }
       },
       pricing: {
         value: {
@@ -15100,9 +15169,50 @@ var catalog_default = {
       unsupportedParameters: [
         "temperature",
         "top_p",
-        "top_k"
+        "top_k",
+        "thinking.type.enabled",
+        "thinking.type.disabled"
       ],
       status: "candidate",
+      deferredToolLoading: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-25T18:30:00Z",
+        sourceRef: 'https://platform.claude.com/docs/en/agents-and-tools/tool-use/tool-search-tool — the Model compatibility table lists Fable 5.1, Mythos 5.1, Fable 5, Mythos 5, Opus 5.5, Opus 5, Opus 4.8, 4.7, 4.6, Sonnet 4.6, Opus 4.5, Sonnet 4.5 and Haiku 4.5 (not Sonnet 5); "Custom tool search implementation": a client tool returns a tool_result whose content carries {type:"tool_reference", tool_name} blocks, every referenced tool declared in tools with defer_loading: true, and the API expands them; "The prefix is untouched, so prompt caching is preserved" (https://platform.claude.com/docs/en/agents-and-tools/tool-use/tool-use-with-prompt-caching).'
+      },
+      midConversationSystem: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-25T19:00:00Z",
+        sourceRef: 'https://platform.claude.com/docs/en/build-with-claude/mid-conversation-system-messages — "This feature is available on Claude Fable 5.1, Claude Mythos 5.1, Claude Fable 5, Claude Mythos 5, Claude Opus 5.5, Claude Opus 4.8, and Claude Opus 5. No beta header is required for mid-conversation system messages. This feature is not available on Claude Sonnet 5."'
+      },
+      midConversationToolChanges: {
+        value: {
+          beta: "mid-conversation-tool-changes-2026-07-01"
+        },
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: 'https://platform.claude.com/docs/en/build-with-claude/mid-conversation-system-messages — tool_addition / tool_removal blocks in a role:"system" message, by reference ("Add/remove by reference: mid-conversation-tool-changes-2026-07-01 — Claude API, Amazon Bedrock, Google Cloud"); available on Claude Fable 5.1, Mythos 5.1, Fable 5, Mythos 5, Opus 5.5, Opus 4.8 and Opus 5, "Not available on Claude Sonnet 5". Retrieved 2026-09-26; release note 2026-07-24 (https://platform.claude.com/docs/en/release-notes/overview).'
+      },
+      inlineToolDefinitions: {
+        value: {
+          beta: "inline-tools-2026-09-15"
+        },
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: 'https://platform.claude.com/docs/en/build-with-claude/mid-conversation-system-messages — tool_addition with a tool_definition ("Add/remove by reference + define by value: inline-tools-2026-09-15 — Claude API"); "The inline-tools-2026-09-15 header covers all reference-based changes"; redefinition: "send a different definition under the same name ... The new definition replaces the earlier one from that position onward" (no removal first, a tool declared in tools included); a definition reusing the name of a different type of tool is a 400 tool_name_conflict. Same model list as midConversationToolChanges. Claude API only, so set on the anthropic and console providers (both api.anthropic.com), never on a cloud reseller. Retrieved 2026-09-26; release note 2026-09-22.'
+      },
+      assistantPrefill: {
+        value: false,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: `https://platform.claude.com/docs/en/models/opus-5-5/migration-guide — "Don't end messages with a prefilled assistant turn: it is rejected"; "Prefilling assistant messages returns a 400 error on Claude Opus 4.6 and later Opus models, including Claude Opus 5.5"; Sonnet 5 and Opus 5: "Manual extended thinking, non-default sampling parameters, and assistant prefill return a 400 error on both models"; the Fable 5.1 guide: prefill "returns a 400 error, unchanged from Claude Fable 5". Live gate 2026-09-26 on claude-opus-5-5: "This model does not support assistant message prefill. The conversation must end with a user message."`
+      },
       canonicalModelId: "claude-fable-5",
       modelFamily: "claude"
     },
@@ -15235,6 +15345,33 @@ var catalog_default = {
           sourceRef: `continuity report §4.4, BOTH halves, applied per anthropic/claude-opus-5's own evidence. Own-state acceptance: a Claude model's own thinking blocks are replayed to it unchanged, in order, with signatures intact. Why the domain is NARROW: prior thinking/redacted_thinking blocks are tied to the model that produced them, so "same provider" is not automatically "same continuation domain" — the domain is this model alone.`,
           confidence: "declared",
           observedAt: "2026-09-07T00:00:00Z"
+        },
+        effortRequest: {
+          value: {
+            field: "output_config.effort"
+          },
+          source: "official-doc",
+          confidence: "declared",
+          observedAt: "2026-09-25T12:30:00Z",
+          sourceRef: "https://platform.claude.com/docs/en/build-with-claude/effort — the effort page's supportedModels list names claude-fable-5-1, claude-fable-5, claude-opus-5-5, claude-opus-5, claude-opus-4-8, claude-opus-4-7, claude-opus-4-6, claude-opus-4-5-20251101, claude-sonnet-5 and claude-sonnet-4-6: effort is requested through `output_config.effort` (Claude 4.7 and later reject a manual `thinking.budget_tokens`, https://platform.claude.com/docs/en/build-with-claude/thinking-troubleshooting#rejected-configurations)."
+        },
+        blockBinding: {
+          value: {
+            beta: "thinking-binding-controls-2026-08-01"
+          },
+          source: "official-doc",
+          confidence: "declared",
+          observedAt: "2026-09-25T13:00:00Z",
+          sourceRef: 'https://platform.claude.com/docs/en/build-with-claude/thinking-troubleshooting — "A 400 error says a thinking block signature is invalid": on Claude Fable 5.1 and Opus 5.5 a replayed thinking block is bound to the conversation (system, tools and earlier messages) and is rejected once that prefix changes (enforced for accounts created on or after 2026-08-31); the documented escape is the `thinking-binding-controls-2026-08-01` beta with `thinking.block_binding.prefix_mismatch_behavior: "drop_block"`. See also https://platform.claude.com/docs/en/models/opus-5-5/whats-new-opus-5-5 ("Thinking blocks are tied to the model and the conversation").'
+        },
+        perMessageEffort: {
+          value: {
+            beta: "mid-conversation-output-config-2026-07-01"
+          },
+          source: "official-doc",
+          confidence: "declared",
+          observedAt: "2026-09-25T18:00:00Z",
+          sourceRef: 'https://platform.claude.com/docs/en/build-with-claude/effort#change-effort-mid-conversation-beta — "On Claude Fable 5.1, Claude Mythos 5.1, Claude Opus 5.5, and Claude Opus 5, use a per-message effort change, which keeps the prompt cache": a role:"system" message with empty content and output_config.effort, beta header mid-conversation-output-config-2026-07-01; "Models without per-message effort, including Claude Fable 5, return a 400 error". claude 2.1.282 sends the older alias per-turn-control-2026-07-01, which the page does not document.'
         }
       },
       pricing: {
@@ -15249,8 +15386,52 @@ var catalog_default = {
         confidence: "declared",
         observedAt: "2026-09-08T00:00:00Z"
       },
-      unsupportedParameters: [],
+      unsupportedParameters: [
+        "thinking.type.enabled",
+        "thinking.type.disabled",
+        "tool_choice.any",
+        "tool_choice.tool"
+      ],
       status: "candidate",
+      deferredToolLoading: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-25T18:30:00Z",
+        sourceRef: 'https://platform.claude.com/docs/en/agents-and-tools/tool-use/tool-search-tool — the Model compatibility table lists Fable 5.1, Mythos 5.1, Fable 5, Mythos 5, Opus 5.5, Opus 5, Opus 4.8, 4.7, 4.6, Sonnet 4.6, Opus 4.5, Sonnet 4.5 and Haiku 4.5 (not Sonnet 5); "Custom tool search implementation": a client tool returns a tool_result whose content carries {type:"tool_reference", tool_name} blocks, every referenced tool declared in tools with defer_loading: true, and the API expands them; "The prefix is untouched, so prompt caching is preserved" (https://platform.claude.com/docs/en/agents-and-tools/tool-use/tool-use-with-prompt-caching).'
+      },
+      midConversationSystem: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-25T19:00:00Z",
+        sourceRef: 'https://platform.claude.com/docs/en/build-with-claude/mid-conversation-system-messages — "This feature is available on Claude Fable 5.1, Claude Mythos 5.1, Claude Fable 5, Claude Mythos 5, Claude Opus 5.5, Claude Opus 4.8, and Claude Opus 5. No beta header is required for mid-conversation system messages. This feature is not available on Claude Sonnet 5."'
+      },
+      midConversationToolChanges: {
+        value: {
+          beta: "mid-conversation-tool-changes-2026-07-01"
+        },
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: 'https://platform.claude.com/docs/en/build-with-claude/mid-conversation-system-messages — tool_addition / tool_removal blocks in a role:"system" message, by reference ("Add/remove by reference: mid-conversation-tool-changes-2026-07-01 — Claude API, Amazon Bedrock, Google Cloud"); available on Claude Fable 5.1, Mythos 5.1, Fable 5, Mythos 5, Opus 5.5, Opus 4.8 and Opus 5, "Not available on Claude Sonnet 5". Retrieved 2026-09-26; release note 2026-07-24 (https://platform.claude.com/docs/en/release-notes/overview).'
+      },
+      inlineToolDefinitions: {
+        value: {
+          beta: "inline-tools-2026-09-15"
+        },
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: 'https://platform.claude.com/docs/en/build-with-claude/mid-conversation-system-messages — tool_addition with a tool_definition ("Add/remove by reference + define by value: inline-tools-2026-09-15 — Claude API"); "The inline-tools-2026-09-15 header covers all reference-based changes"; redefinition: "send a different definition under the same name ... The new definition replaces the earlier one from that position onward" (no removal first, a tool declared in tools included); a definition reusing the name of a different type of tool is a 400 tool_name_conflict. Same model list as midConversationToolChanges. Claude API only, so set on the anthropic and console providers (both api.anthropic.com), never on a cloud reseller. Retrieved 2026-09-26; release note 2026-09-22.'
+      },
+      assistantPrefill: {
+        value: false,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: `https://platform.claude.com/docs/en/models/opus-5-5/migration-guide — "Don't end messages with a prefilled assistant turn: it is rejected"; "Prefilling assistant messages returns a 400 error on Claude Opus 4.6 and later Opus models, including Claude Opus 5.5"; Sonnet 5 and Opus 5: "Manual extended thinking, non-default sampling parameters, and assistant prefill return a 400 error on both models"; the Fable 5.1 guide: prefill "returns a 400 error, unchanged from Claude Fable 5". Live gate 2026-09-26 on claude-opus-5-5: "This model does not support assistant message prefill. The conversation must end with a user message."`
+      },
       canonicalModelId: "claude-fable-5.1",
       modelFamily: "claude"
     },
@@ -15345,8 +15526,17 @@ var catalog_default = {
         confidence: "declared",
         observedAt: "2026-09-05T00:00:00Z"
       },
-      unsupportedParameters: [],
+      unsupportedParameters: [
+        "thinking.type.adaptive"
+      ],
       status: "candidate",
+      deferredToolLoading: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-25T18:30:00Z",
+        sourceRef: 'https://platform.claude.com/docs/en/agents-and-tools/tool-use/tool-search-tool — the Model compatibility table lists Fable 5.1, Mythos 5.1, Fable 5, Mythos 5, Opus 5.5, Opus 5, Opus 4.8, 4.7, 4.6, Sonnet 4.6, Opus 4.5, Sonnet 4.5 and Haiku 4.5 (not Sonnet 5); "Custom tool search implementation": a client tool returns a tool_result whose content carries {type:"tool_reference", tool_name} blocks, every referenced tool declared in tools with defer_loading: true, and the API expands them; "The prefix is untouched, so prompt caching is preserved" (https://platform.claude.com/docs/en/agents-and-tools/tool-use/tool-use-with-prompt-caching).'
+      },
       canonicalModelId: "claude-haiku-4.5-20251001",
       modelFamily: "claude"
     },
@@ -15441,8 +15631,17 @@ var catalog_default = {
         confidence: "declared",
         observedAt: "2026-09-19T00:00:00Z"
       },
-      unsupportedParameters: [],
+      unsupportedParameters: [
+        "thinking.type.adaptive"
+      ],
       status: "candidate",
+      deferredToolLoading: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-25T18:30:00Z",
+        sourceRef: 'https://platform.claude.com/docs/en/agents-and-tools/tool-use/tool-search-tool — the Model compatibility table lists Fable 5.1, Mythos 5.1, Fable 5, Mythos 5, Opus 5.5, Opus 5, Opus 4.8, 4.7, 4.6, Sonnet 4.6, Opus 4.5, Sonnet 4.5 and Haiku 4.5 (not Sonnet 5); "Custom tool search implementation": a client tool returns a tool_result whose content carries {type:"tool_reference", tool_name} blocks, every referenced tool declared in tools with defer_loading: true, and the API expands them; "The prefix is untouched, so prompt caching is preserved" (https://platform.claude.com/docs/en/agents-and-tools/tool-use/tool-use-with-prompt-caching).'
+      },
       canonicalModelId: "claude-haiku-4.5",
       modelFamily: "claude"
     },
@@ -15540,7 +15739,16 @@ var catalog_default = {
           "high"
         ],
         continuation: "opaque-provider-state",
-        defaultEffort: "high"
+        defaultEffort: "high",
+        effortRequest: {
+          value: {
+            field: "output_config.effort"
+          },
+          source: "official-doc",
+          confidence: "declared",
+          observedAt: "2026-09-25T12:30:00Z",
+          sourceRef: "https://platform.claude.com/docs/en/build-with-claude/effort — the effort page's supportedModels list names claude-fable-5-1, claude-fable-5, claude-opus-5-5, claude-opus-5, claude-opus-4-8, claude-opus-4-7, claude-opus-4-6, claude-opus-4-5-20251101, claude-sonnet-5 and claude-sonnet-4-6: effort is requested through `output_config.effort` (Claude 4.7 and later reject a manual `thinking.budget_tokens`, https://platform.claude.com/docs/en/build-with-claude/thinking-troubleshooting#rejected-configurations)."
+        }
       },
       pricing: {
         value: {
@@ -15554,8 +15762,17 @@ var catalog_default = {
         confidence: "declared",
         observedAt: "2026-09-19T00:00:00Z"
       },
-      unsupportedParameters: [],
+      unsupportedParameters: [
+        "thinking.type.adaptive"
+      ],
       status: "candidate",
+      deferredToolLoading: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-25T18:30:00Z",
+        sourceRef: 'https://platform.claude.com/docs/en/agents-and-tools/tool-use/tool-search-tool — the Model compatibility table lists Fable 5.1, Mythos 5.1, Fable 5, Mythos 5, Opus 5.5, Opus 5, Opus 4.8, 4.7, 4.6, Sonnet 4.6, Opus 4.5, Sonnet 4.5 and Haiku 4.5 (not Sonnet 5); "Custom tool search implementation": a client tool returns a tool_result whose content carries {type:"tool_reference", tool_name} blocks, every referenced tool declared in tools with defer_loading: true, and the API expands them; "The prefix is untouched, so prompt caching is preserved" (https://platform.claude.com/docs/en/agents-and-tools/tool-use/tool-use-with-prompt-caching).'
+      },
       canonicalModelId: "claude-opus-4.5",
       modelFamily: "claude"
     },
@@ -15564,7 +15781,9 @@ var catalog_default = {
       providerId: "anthropic",
       upstreamId: "claude-opus-4.6",
       displayName: "Claude Opus 4.6",
-      aliases: [],
+      aliases: [
+        "claude-opus-4-6"
+      ],
       endpoints: [
         "chat"
       ],
@@ -15651,7 +15870,16 @@ var catalog_default = {
           "max"
         ],
         continuation: "opaque-provider-state",
-        defaultEffort: "high"
+        defaultEffort: "high",
+        effortRequest: {
+          value: {
+            field: "output_config.effort"
+          },
+          source: "official-doc",
+          confidence: "declared",
+          observedAt: "2026-09-25T12:30:00Z",
+          sourceRef: "https://platform.claude.com/docs/en/build-with-claude/effort — the effort page's supportedModels list names claude-fable-5-1, claude-fable-5, claude-opus-5-5, claude-opus-5, claude-opus-4-8, claude-opus-4-7, claude-opus-4-6, claude-opus-4-5-20251101, claude-sonnet-5 and claude-sonnet-4-6: effort is requested through `output_config.effort` (Claude 4.7 and later reject a manual `thinking.budget_tokens`, https://platform.claude.com/docs/en/build-with-claude/thinking-troubleshooting#rejected-configurations)."
+        }
       },
       pricing: {
         value: {
@@ -15667,6 +15895,20 @@ var catalog_default = {
       },
       unsupportedParameters: [],
       status: "candidate",
+      deferredToolLoading: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-25T18:30:00Z",
+        sourceRef: 'https://platform.claude.com/docs/en/agents-and-tools/tool-use/tool-search-tool — the Model compatibility table lists Fable 5.1, Mythos 5.1, Fable 5, Mythos 5, Opus 5.5, Opus 5, Opus 4.8, 4.7, 4.6, Sonnet 4.6, Opus 4.5, Sonnet 4.5 and Haiku 4.5 (not Sonnet 5); "Custom tool search implementation": a client tool returns a tool_result whose content carries {type:"tool_reference", tool_name} blocks, every referenced tool declared in tools with defer_loading: true, and the API expands them; "The prefix is untouched, so prompt caching is preserved" (https://platform.claude.com/docs/en/agents-and-tools/tool-use/tool-use-with-prompt-caching).'
+      },
+      assistantPrefill: {
+        value: false,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: `https://platform.claude.com/docs/en/models/opus-5-5/migration-guide — "Don't end messages with a prefilled assistant turn: it is rejected"; "Prefilling assistant messages returns a 400 error on Claude Opus 4.6 and later Opus models, including Claude Opus 5.5"; Sonnet 5 and Opus 5: "Manual extended thinking, non-default sampling parameters, and assistant prefill return a 400 error on both models"; the Fable 5.1 guide: prefill "returns a 400 error, unchanged from Claude Fable 5". Live gate 2026-09-26 on claude-opus-5-5: "This model does not support assistant message prefill. The conversation must end with a user message."`
+      },
       canonicalModelId: "claude-opus-4.6",
       modelFamily: "claude"
     },
@@ -15675,7 +15917,9 @@ var catalog_default = {
       providerId: "anthropic",
       upstreamId: "claude-opus-4.7",
       displayName: "Claude Opus 4.7",
-      aliases: [],
+      aliases: [
+        "claude-opus-4-7"
+      ],
       endpoints: [
         "chat"
       ],
@@ -15763,7 +16007,16 @@ var catalog_default = {
           "max"
         ],
         continuation: "opaque-provider-state",
-        defaultEffort: "high"
+        defaultEffort: "high",
+        effortRequest: {
+          value: {
+            field: "output_config.effort"
+          },
+          source: "official-doc",
+          confidence: "declared",
+          observedAt: "2026-09-25T12:30:00Z",
+          sourceRef: "https://platform.claude.com/docs/en/build-with-claude/effort — the effort page's supportedModels list names claude-fable-5-1, claude-fable-5, claude-opus-5-5, claude-opus-5, claude-opus-4-8, claude-opus-4-7, claude-opus-4-6, claude-opus-4-5-20251101, claude-sonnet-5 and claude-sonnet-4-6: effort is requested through `output_config.effort` (Claude 4.7 and later reject a manual `thinking.budget_tokens`, https://platform.claude.com/docs/en/build-with-claude/thinking-troubleshooting#rejected-configurations)."
+        }
       },
       pricing: {
         value: {
@@ -15780,9 +16033,24 @@ var catalog_default = {
       unsupportedParameters: [
         "temperature",
         "top_p",
-        "top_k"
+        "top_k",
+        "thinking.type.enabled"
       ],
       status: "candidate",
+      deferredToolLoading: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-25T18:30:00Z",
+        sourceRef: 'https://platform.claude.com/docs/en/agents-and-tools/tool-use/tool-search-tool — the Model compatibility table lists Fable 5.1, Mythos 5.1, Fable 5, Mythos 5, Opus 5.5, Opus 5, Opus 4.8, 4.7, 4.6, Sonnet 4.6, Opus 4.5, Sonnet 4.5 and Haiku 4.5 (not Sonnet 5); "Custom tool search implementation": a client tool returns a tool_result whose content carries {type:"tool_reference", tool_name} blocks, every referenced tool declared in tools with defer_loading: true, and the API expands them; "The prefix is untouched, so prompt caching is preserved" (https://platform.claude.com/docs/en/agents-and-tools/tool-use/tool-use-with-prompt-caching).'
+      },
+      assistantPrefill: {
+        value: false,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: `https://platform.claude.com/docs/en/models/opus-5-5/migration-guide — "Don't end messages with a prefilled assistant turn: it is rejected"; "Prefilling assistant messages returns a 400 error on Claude Opus 4.6 and later Opus models, including Claude Opus 5.5"; Sonnet 5 and Opus 5: "Manual extended thinking, non-default sampling parameters, and assistant prefill return a 400 error on both models"; the Fable 5.1 guide: prefill "returns a 400 error, unchanged from Claude Fable 5". Live gate 2026-09-26 on claude-opus-5-5: "This model does not support assistant message prefill. The conversation must end with a user message."`
+      },
       canonicalModelId: "claude-opus-4.7",
       modelFamily: "claude"
     },
@@ -15791,7 +16059,9 @@ var catalog_default = {
       providerId: "anthropic",
       upstreamId: "claude-opus-4.8",
       displayName: "Claude Opus 4.8",
-      aliases: [],
+      aliases: [
+        "claude-opus-4-8"
+      ],
       endpoints: [
         "chat"
       ],
@@ -15879,7 +16149,16 @@ var catalog_default = {
           "max"
         ],
         continuation: "opaque-provider-state",
-        defaultEffort: "high"
+        defaultEffort: "high",
+        effortRequest: {
+          value: {
+            field: "output_config.effort"
+          },
+          source: "official-doc",
+          confidence: "declared",
+          observedAt: "2026-09-25T12:30:00Z",
+          sourceRef: "https://platform.claude.com/docs/en/build-with-claude/effort — the effort page's supportedModels list names claude-fable-5-1, claude-fable-5, claude-opus-5-5, claude-opus-5, claude-opus-4-8, claude-opus-4-7, claude-opus-4-6, claude-opus-4-5-20251101, claude-sonnet-5 and claude-sonnet-4-6: effort is requested through `output_config.effort` (Claude 4.7 and later reject a manual `thinking.budget_tokens`, https://platform.claude.com/docs/en/build-with-claude/thinking-troubleshooting#rejected-configurations)."
+        }
       },
       pricing: {
         value: {
@@ -15896,9 +16175,49 @@ var catalog_default = {
       unsupportedParameters: [
         "temperature",
         "top_p",
-        "top_k"
+        "top_k",
+        "thinking.type.enabled"
       ],
       status: "candidate",
+      deferredToolLoading: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-25T18:30:00Z",
+        sourceRef: 'https://platform.claude.com/docs/en/agents-and-tools/tool-use/tool-search-tool — the Model compatibility table lists Fable 5.1, Mythos 5.1, Fable 5, Mythos 5, Opus 5.5, Opus 5, Opus 4.8, 4.7, 4.6, Sonnet 4.6, Opus 4.5, Sonnet 4.5 and Haiku 4.5 (not Sonnet 5); "Custom tool search implementation": a client tool returns a tool_result whose content carries {type:"tool_reference", tool_name} blocks, every referenced tool declared in tools with defer_loading: true, and the API expands them; "The prefix is untouched, so prompt caching is preserved" (https://platform.claude.com/docs/en/agents-and-tools/tool-use/tool-use-with-prompt-caching).'
+      },
+      midConversationSystem: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-25T19:00:00Z",
+        sourceRef: 'https://platform.claude.com/docs/en/build-with-claude/mid-conversation-system-messages — "This feature is available on Claude Fable 5.1, Claude Mythos 5.1, Claude Fable 5, Claude Mythos 5, Claude Opus 5.5, Claude Opus 4.8, and Claude Opus 5. No beta header is required for mid-conversation system messages. This feature is not available on Claude Sonnet 5."'
+      },
+      midConversationToolChanges: {
+        value: {
+          beta: "mid-conversation-tool-changes-2026-07-01"
+        },
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: 'https://platform.claude.com/docs/en/build-with-claude/mid-conversation-system-messages — tool_addition / tool_removal blocks in a role:"system" message, by reference ("Add/remove by reference: mid-conversation-tool-changes-2026-07-01 — Claude API, Amazon Bedrock, Google Cloud"); available on Claude Fable 5.1, Mythos 5.1, Fable 5, Mythos 5, Opus 5.5, Opus 4.8 and Opus 5, "Not available on Claude Sonnet 5". Retrieved 2026-09-26; release note 2026-07-24 (https://platform.claude.com/docs/en/release-notes/overview).'
+      },
+      inlineToolDefinitions: {
+        value: {
+          beta: "inline-tools-2026-09-15"
+        },
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: 'https://platform.claude.com/docs/en/build-with-claude/mid-conversation-system-messages — tool_addition with a tool_definition ("Add/remove by reference + define by value: inline-tools-2026-09-15 — Claude API"); "The inline-tools-2026-09-15 header covers all reference-based changes"; redefinition: "send a different definition under the same name ... The new definition replaces the earlier one from that position onward" (no removal first, a tool declared in tools included); a definition reusing the name of a different type of tool is a 400 tool_name_conflict. Same model list as midConversationToolChanges. Claude API only, so set on the anthropic and console providers (both api.anthropic.com), never on a cloud reseller. Retrieved 2026-09-26; release note 2026-09-22.'
+      },
+      assistantPrefill: {
+        value: false,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: `https://platform.claude.com/docs/en/models/opus-5-5/migration-guide — "Don't end messages with a prefilled assistant turn: it is rejected"; "Prefilling assistant messages returns a 400 error on Claude Opus 4.6 and later Opus models, including Claude Opus 5.5"; Sonnet 5 and Opus 5: "Manual extended thinking, non-default sampling parameters, and assistant prefill return a 400 error on both models"; the Fable 5.1 guide: prefill "returns a 400 error, unchanged from Claude Fable 5". Live gate 2026-09-26 on claude-opus-5-5: "This model does not support assistant message prefill. The conversation must end with a user message."`
+      },
       canonicalModelId: "claude-opus-4.8",
       modelFamily: "claude"
     },
@@ -16038,6 +16357,24 @@ var catalog_default = {
           sourceRef: "continuity report §4.4, BOTH halves. Own-state acceptance: the model's own `thinking` blocks are replayed to it unchanged, in order, with their signatures intact (§4.4's replay rules and its worked request). Why the domain is NARROW: \"when changing Claude models, prior `thinking` and `redacted_thinking` blocks should be stripped because they are tied to the model that produced them. Therefore 'same provider' is not automatically 'same continuation domain.'\" The domain is this model alone, never the Anthropic provider.",
           confidence: "declared",
           observedAt: "2026-09-05T00:00:00Z"
+        },
+        effortRequest: {
+          value: {
+            field: "output_config.effort"
+          },
+          source: "official-doc",
+          confidence: "declared",
+          observedAt: "2026-09-25T12:30:00Z",
+          sourceRef: "https://platform.claude.com/docs/en/build-with-claude/effort — the effort page's supportedModels list names claude-fable-5-1, claude-fable-5, claude-opus-5-5, claude-opus-5, claude-opus-4-8, claude-opus-4-7, claude-opus-4-6, claude-opus-4-5-20251101, claude-sonnet-5 and claude-sonnet-4-6: effort is requested through `output_config.effort` (Claude 4.7 and later reject a manual `thinking.budget_tokens`, https://platform.claude.com/docs/en/build-with-claude/thinking-troubleshooting#rejected-configurations)."
+        },
+        perMessageEffort: {
+          value: {
+            beta: "mid-conversation-output-config-2026-07-01"
+          },
+          source: "official-doc",
+          confidence: "declared",
+          observedAt: "2026-09-25T18:00:00Z",
+          sourceRef: 'https://platform.claude.com/docs/en/build-with-claude/effort#change-effort-mid-conversation-beta — "On Claude Fable 5.1, Claude Mythos 5.1, Claude Opus 5.5, and Claude Opus 5, use a per-message effort change, which keeps the prompt cache": a role:"system" message with empty content and output_config.effort, beta header mid-conversation-output-config-2026-07-01; "Models without per-message effort, including Claude Fable 5, return a 400 error". claude 2.1.282 sends the older alias per-turn-control-2026-07-01, which the page does not document.'
         }
       },
       pricing: {
@@ -16055,9 +16392,51 @@ var catalog_default = {
       unsupportedParameters: [
         "temperature",
         "top_p",
-        "top_k"
+        "top_k",
+        "thinking.type.enabled",
+        "thinking.type.disabled+output_config.effort.xhigh",
+        "thinking.type.disabled+output_config.effort.max"
       ],
       status: "candidate",
+      deferredToolLoading: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-25T18:30:00Z",
+        sourceRef: 'https://platform.claude.com/docs/en/agents-and-tools/tool-use/tool-search-tool — the Model compatibility table lists Fable 5.1, Mythos 5.1, Fable 5, Mythos 5, Opus 5.5, Opus 5, Opus 4.8, 4.7, 4.6, Sonnet 4.6, Opus 4.5, Sonnet 4.5 and Haiku 4.5 (not Sonnet 5); "Custom tool search implementation": a client tool returns a tool_result whose content carries {type:"tool_reference", tool_name} blocks, every referenced tool declared in tools with defer_loading: true, and the API expands them; "The prefix is untouched, so prompt caching is preserved" (https://platform.claude.com/docs/en/agents-and-tools/tool-use/tool-use-with-prompt-caching).'
+      },
+      midConversationSystem: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-25T19:00:00Z",
+        sourceRef: 'https://platform.claude.com/docs/en/build-with-claude/mid-conversation-system-messages — "This feature is available on Claude Fable 5.1, Claude Mythos 5.1, Claude Fable 5, Claude Mythos 5, Claude Opus 5.5, Claude Opus 4.8, and Claude Opus 5. No beta header is required for mid-conversation system messages. This feature is not available on Claude Sonnet 5."'
+      },
+      midConversationToolChanges: {
+        value: {
+          beta: "mid-conversation-tool-changes-2026-07-01"
+        },
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: 'https://platform.claude.com/docs/en/build-with-claude/mid-conversation-system-messages — tool_addition / tool_removal blocks in a role:"system" message, by reference ("Add/remove by reference: mid-conversation-tool-changes-2026-07-01 — Claude API, Amazon Bedrock, Google Cloud"); available on Claude Fable 5.1, Mythos 5.1, Fable 5, Mythos 5, Opus 5.5, Opus 4.8 and Opus 5, "Not available on Claude Sonnet 5". Retrieved 2026-09-26; release note 2026-07-24 (https://platform.claude.com/docs/en/release-notes/overview).'
+      },
+      inlineToolDefinitions: {
+        value: {
+          beta: "inline-tools-2026-09-15"
+        },
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: 'https://platform.claude.com/docs/en/build-with-claude/mid-conversation-system-messages — tool_addition with a tool_definition ("Add/remove by reference + define by value: inline-tools-2026-09-15 — Claude API"); "The inline-tools-2026-09-15 header covers all reference-based changes"; redefinition: "send a different definition under the same name ... The new definition replaces the earlier one from that position onward" (no removal first, a tool declared in tools included); a definition reusing the name of a different type of tool is a 400 tool_name_conflict. Same model list as midConversationToolChanges. Claude API only, so set on the anthropic and console providers (both api.anthropic.com), never on a cloud reseller. Retrieved 2026-09-26; release note 2026-09-22.'
+      },
+      assistantPrefill: {
+        value: false,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: `https://platform.claude.com/docs/en/models/opus-5-5/migration-guide — "Don't end messages with a prefilled assistant turn: it is rejected"; "Prefilling assistant messages returns a 400 error on Claude Opus 4.6 and later Opus models, including Claude Opus 5.5"; Sonnet 5 and Opus 5: "Manual extended thinking, non-default sampling parameters, and assistant prefill return a 400 error on both models"; the Fable 5.1 guide: prefill "returns a 400 error, unchanged from Claude Fable 5". Live gate 2026-09-26 on claude-opus-5-5: "This model does not support assistant message prefill. The conversation must end with a user message."`
+      },
       canonicalModelId: "claude-opus-5",
       modelFamily: "claude"
     },
@@ -16199,6 +16578,33 @@ var catalog_default = {
           sourceRef: "continuity report §3 (wire probe of the pinned Claude Agent SDK, 0.3.250→0.3.258) — every uncompacted historical thinking/redacted block is replayed on every later request",
           confidence: "declared",
           observedAt: "2026-09-05T00:00:00Z"
+        },
+        effortRequest: {
+          value: {
+            field: "output_config.effort"
+          },
+          source: "official-doc",
+          confidence: "declared",
+          observedAt: "2026-09-25T12:30:00Z",
+          sourceRef: "https://platform.claude.com/docs/en/build-with-claude/effort — the effort page's supportedModels list names claude-fable-5-1, claude-fable-5, claude-opus-5-5, claude-opus-5, claude-opus-4-8, claude-opus-4-7, claude-opus-4-6, claude-opus-4-5-20251101, claude-sonnet-5 and claude-sonnet-4-6: effort is requested through `output_config.effort` (Claude 4.7 and later reject a manual `thinking.budget_tokens`, https://platform.claude.com/docs/en/build-with-claude/thinking-troubleshooting#rejected-configurations)."
+        },
+        blockBinding: {
+          value: {
+            beta: "thinking-binding-controls-2026-08-01"
+          },
+          source: "official-doc",
+          confidence: "declared",
+          observedAt: "2026-09-25T13:00:00Z",
+          sourceRef: 'https://platform.claude.com/docs/en/build-with-claude/thinking-troubleshooting — "A 400 error says a thinking block signature is invalid": on Claude Fable 5.1 and Opus 5.5 a replayed thinking block is bound to the conversation (system, tools and earlier messages) and is rejected once that prefix changes (enforced for accounts created on or after 2026-08-31); the documented escape is the `thinking-binding-controls-2026-08-01` beta with `thinking.block_binding.prefix_mismatch_behavior: "drop_block"`. See also https://platform.claude.com/docs/en/models/opus-5-5/whats-new-opus-5-5 ("Thinking blocks are tied to the model and the conversation").'
+        },
+        perMessageEffort: {
+          value: {
+            beta: "mid-conversation-output-config-2026-07-01"
+          },
+          source: "official-doc",
+          confidence: "declared",
+          observedAt: "2026-09-25T18:00:00Z",
+          sourceRef: 'https://platform.claude.com/docs/en/build-with-claude/effort#change-effort-mid-conversation-beta — "On Claude Fable 5.1, Claude Mythos 5.1, Claude Opus 5.5, and Claude Opus 5, use a per-message effort change, which keeps the prompt cache": a role:"system" message with empty content and output_config.effort, beta header mid-conversation-output-config-2026-07-01; "Models without per-message effort, including Claude Fable 5, return a 400 error". claude 2.1.282 sends the older alias per-turn-control-2026-07-01, which the page does not document.'
         }
       },
       pricing: {
@@ -16213,8 +16619,52 @@ var catalog_default = {
         confidence: "declared",
         observedAt: "2026-09-25T10:25:20Z"
       },
-      unsupportedParameters: [],
+      unsupportedParameters: [
+        "thinking.type.enabled",
+        "thinking.type.disabled",
+        "tool_choice.any",
+        "tool_choice.tool"
+      ],
       status: "candidate",
+      deferredToolLoading: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-25T18:30:00Z",
+        sourceRef: 'https://platform.claude.com/docs/en/agents-and-tools/tool-use/tool-search-tool — the Model compatibility table lists Fable 5.1, Mythos 5.1, Fable 5, Mythos 5, Opus 5.5, Opus 5, Opus 4.8, 4.7, 4.6, Sonnet 4.6, Opus 4.5, Sonnet 4.5 and Haiku 4.5 (not Sonnet 5); "Custom tool search implementation": a client tool returns a tool_result whose content carries {type:"tool_reference", tool_name} blocks, every referenced tool declared in tools with defer_loading: true, and the API expands them; "The prefix is untouched, so prompt caching is preserved" (https://platform.claude.com/docs/en/agents-and-tools/tool-use/tool-use-with-prompt-caching).'
+      },
+      midConversationSystem: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-25T19:00:00Z",
+        sourceRef: 'https://platform.claude.com/docs/en/build-with-claude/mid-conversation-system-messages — "This feature is available on Claude Fable 5.1, Claude Mythos 5.1, Claude Fable 5, Claude Mythos 5, Claude Opus 5.5, Claude Opus 4.8, and Claude Opus 5. No beta header is required for mid-conversation system messages. This feature is not available on Claude Sonnet 5."'
+      },
+      midConversationToolChanges: {
+        value: {
+          beta: "mid-conversation-tool-changes-2026-07-01"
+        },
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: 'https://platform.claude.com/docs/en/build-with-claude/mid-conversation-system-messages — tool_addition / tool_removal blocks in a role:"system" message, by reference ("Add/remove by reference: mid-conversation-tool-changes-2026-07-01 — Claude API, Amazon Bedrock, Google Cloud"); available on Claude Fable 5.1, Mythos 5.1, Fable 5, Mythos 5, Opus 5.5, Opus 4.8 and Opus 5, "Not available on Claude Sonnet 5". Retrieved 2026-09-26; release note 2026-07-24 (https://platform.claude.com/docs/en/release-notes/overview).'
+      },
+      inlineToolDefinitions: {
+        value: {
+          beta: "inline-tools-2026-09-15"
+        },
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: 'https://platform.claude.com/docs/en/build-with-claude/mid-conversation-system-messages — tool_addition with a tool_definition ("Add/remove by reference + define by value: inline-tools-2026-09-15 — Claude API"); "The inline-tools-2026-09-15 header covers all reference-based changes"; redefinition: "send a different definition under the same name ... The new definition replaces the earlier one from that position onward" (no removal first, a tool declared in tools included); a definition reusing the name of a different type of tool is a 400 tool_name_conflict. Same model list as midConversationToolChanges. Claude API only, so set on the anthropic and console providers (both api.anthropic.com), never on a cloud reseller. Retrieved 2026-09-26; release note 2026-09-22.'
+      },
+      assistantPrefill: {
+        value: false,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: `https://platform.claude.com/docs/en/models/opus-5-5/migration-guide — "Don't end messages with a prefilled assistant turn: it is rejected"; "Prefilling assistant messages returns a 400 error on Claude Opus 4.6 and later Opus models, including Claude Opus 5.5"; Sonnet 5 and Opus 5: "Manual extended thinking, non-default sampling parameters, and assistant prefill return a 400 error on both models"; the Fable 5.1 guide: prefill "returns a 400 error, unchanged from Claude Fable 5". Live gate 2026-09-26 on claude-opus-5-5: "This model does not support assistant message prefill. The conversation must end with a user message."`
+      },
       canonicalModelId: "claude-opus-5.5",
       modelFamily: "claude"
     },
@@ -16326,8 +16776,17 @@ var catalog_default = {
         confidence: "declared",
         observedAt: "2026-09-19T00:00:00Z"
       },
-      unsupportedParameters: [],
+      unsupportedParameters: [
+        "thinking.type.adaptive"
+      ],
       status: "candidate",
+      deferredToolLoading: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-25T18:30:00Z",
+        sourceRef: 'https://platform.claude.com/docs/en/agents-and-tools/tool-use/tool-search-tool — the Model compatibility table lists Fable 5.1, Mythos 5.1, Fable 5, Mythos 5, Opus 5.5, Opus 5, Opus 4.8, 4.7, 4.6, Sonnet 4.6, Opus 4.5, Sonnet 4.5 and Haiku 4.5 (not Sonnet 5); "Custom tool search implementation": a client tool returns a tool_result whose content carries {type:"tool_reference", tool_name} blocks, every referenced tool declared in tools with defer_loading: true, and the API expands them; "The prefix is untouched, so prompt caching is preserved" (https://platform.claude.com/docs/en/agents-and-tools/tool-use/tool-use-with-prompt-caching).'
+      },
       canonicalModelId: "claude-sonnet-4.5",
       modelFamily: "claude"
     },
@@ -16336,7 +16795,9 @@ var catalog_default = {
       providerId: "anthropic",
       upstreamId: "claude-sonnet-4.6",
       displayName: "Claude Sonnet 4.6",
-      aliases: [],
+      aliases: [
+        "claude-sonnet-4-6"
+      ],
       endpoints: [
         "chat"
       ],
@@ -16423,7 +16884,16 @@ var catalog_default = {
           "max"
         ],
         continuation: "opaque-provider-state",
-        defaultEffort: "high"
+        defaultEffort: "high",
+        effortRequest: {
+          value: {
+            field: "output_config.effort"
+          },
+          source: "official-doc",
+          confidence: "declared",
+          observedAt: "2026-09-25T12:30:00Z",
+          sourceRef: "https://platform.claude.com/docs/en/build-with-claude/effort — the effort page's supportedModels list names claude-fable-5-1, claude-fable-5, claude-opus-5-5, claude-opus-5, claude-opus-4-8, claude-opus-4-7, claude-opus-4-6, claude-opus-4-5-20251101, claude-sonnet-5 and claude-sonnet-4-6: effort is requested through `output_config.effort` (Claude 4.7 and later reject a manual `thinking.budget_tokens`, https://platform.claude.com/docs/en/build-with-claude/thinking-troubleshooting#rejected-configurations)."
+        }
       },
       pricing: {
         value: {
@@ -16439,6 +16909,13 @@ var catalog_default = {
       },
       unsupportedParameters: [],
       status: "candidate",
+      deferredToolLoading: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-25T18:30:00Z",
+        sourceRef: 'https://platform.claude.com/docs/en/agents-and-tools/tool-use/tool-search-tool — the Model compatibility table lists Fable 5.1, Mythos 5.1, Fable 5, Mythos 5, Opus 5.5, Opus 5, Opus 4.8, 4.7, 4.6, Sonnet 4.6, Opus 4.5, Sonnet 4.5 and Haiku 4.5 (not Sonnet 5); "Custom tool search implementation": a client tool returns a tool_result whose content carries {type:"tool_reference", tool_name} blocks, every referenced tool declared in tools with defer_loading: true, and the API expands them; "The prefix is untouched, so prompt caching is preserved" (https://platform.claude.com/docs/en/agents-and-tools/tool-use/tool-use-with-prompt-caching).'
+      },
       canonicalModelId: "claude-sonnet-4.6",
       modelFamily: "claude"
     },
@@ -16580,6 +17057,15 @@ var catalog_default = {
           sourceRef: 'continuity report §4.4, BOTH halves. Own-state acceptance: §4.4 requires the model\'s own assistant blocks — `thinking` with its signature, `redacted_thinking` — replayed unchanged and in order across a tool loop. Why the domain is NARROW: Anthropic documents model switching as a boundary at which those blocks are STRIPPED, because they are tied to the producing model; "same provider" is therefore not "same continuation domain", and this domain is the model alone.',
           confidence: "declared",
           observedAt: "2026-09-05T00:00:00Z"
+        },
+        effortRequest: {
+          value: {
+            field: "output_config.effort"
+          },
+          source: "official-doc",
+          confidence: "declared",
+          observedAt: "2026-09-25T12:30:00Z",
+          sourceRef: "https://platform.claude.com/docs/en/build-with-claude/effort — the effort page's supportedModels list names claude-fable-5-1, claude-fable-5, claude-opus-5-5, claude-opus-5, claude-opus-4-8, claude-opus-4-7, claude-opus-4-6, claude-opus-4-5-20251101, claude-sonnet-5 and claude-sonnet-4-6: effort is requested through `output_config.effort` (Claude 4.7 and later reject a manual `thinking.budget_tokens`, https://platform.claude.com/docs/en/build-with-claude/thinking-troubleshooting#rejected-configurations)."
         }
       },
       pricing: {
@@ -16597,9 +17083,17 @@ var catalog_default = {
       unsupportedParameters: [
         "temperature",
         "top_p",
-        "top_k"
+        "top_k",
+        "thinking.type.enabled"
       ],
       status: "candidate",
+      assistantPrefill: {
+        value: false,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: `https://platform.claude.com/docs/en/models/opus-5-5/migration-guide — "Don't end messages with a prefilled assistant turn: it is rejected"; "Prefilling assistant messages returns a 400 error on Claude Opus 4.6 and later Opus models, including Claude Opus 5.5"; Sonnet 5 and Opus 5: "Manual extended thinking, non-default sampling parameters, and assistant prefill return a 400 error on both models"; the Fable 5.1 guide: prefill "returns a 400 error, unchanged from Claude Fable 5". Live gate 2026-09-26 on claude-opus-5-5: "This model does not support assistant message prefill. The conversation must end with a user message."`
+      },
       canonicalModelId: "claude-sonnet-5",
       modelFamily: "claude"
     },
@@ -22444,13 +22938,12 @@ var catalog_default = {
         value: [
           "text",
           "image",
-          "video",
-          "pdf"
+          "video"
         ],
         source: "official-doc",
-        sourceRef: "https://docs.aws.amazon.com/nova/latest/userguide/what-is-nova.html — Amazon Nova Lite supports text, image, video, pdf inputs; PDF is mapped from document support",
-        confidence: "inferred",
-        observedAt: "2026-09-25T11:00:00Z"
+        sourceRef: "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-amazon-nova-lite.html — per-model Input Modalities table marks Text, Image, Video supported; PDF is not listed as a separate modality",
+        confidence: "declared",
+        observedAt: "2026-09-25T12:31:56.381Z"
       },
       outputModalities: {
         value: [
@@ -22595,13 +23088,12 @@ var catalog_default = {
         value: [
           "text",
           "image",
-          "video",
-          "pdf"
+          "video"
         ],
         source: "official-doc",
-        sourceRef: "https://docs.aws.amazon.com/nova/latest/userguide/what-is-nova.html — Amazon Nova Premier supports text, image, video, pdf inputs; PDF is mapped from document support",
-        confidence: "inferred",
-        observedAt: "2026-09-25T11:00:00Z"
+        sourceRef: "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-amazon-nova-premier.html — per-model Input Modalities table marks Text, Image, Video supported; PDF is not listed as a separate modality",
+        confidence: "declared",
+        observedAt: "2026-09-25T12:31:56.381Z"
       },
       outputModalities: {
         value: [
@@ -22683,13 +23175,12 @@ var catalog_default = {
         value: [
           "text",
           "image",
-          "video",
-          "pdf"
+          "video"
         ],
         source: "official-doc",
-        sourceRef: "https://docs.aws.amazon.com/nova/latest/userguide/what-is-nova.html — Amazon Nova Pro supports text, image, video, pdf inputs; PDF is mapped from document support",
-        confidence: "inferred",
-        observedAt: "2026-09-25T11:00:00Z"
+        sourceRef: "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-amazon-nova-pro.html — per-model Input Modalities table marks Text, Image, Video supported; PDF is not listed as a separate modality",
+        confidence: "declared",
+        observedAt: "2026-09-25T12:31:56.381Z"
       },
       outputModalities: {
         value: [
@@ -23127,13 +23618,12 @@ var catalog_default = {
         value: [
           "text",
           "image",
-          "video",
-          "pdf"
+          "video"
         ],
         source: "official-doc",
-        sourceRef: "https://docs.aws.amazon.com/nova/latest/nova2-userguide/what-is-nova-2.html — Amazon Nova 2 Lite (US Geo) supports text, image, video, pdf inputs; PDF is mapped from document support",
-        confidence: "inferred",
-        observedAt: "2026-09-25T11:00:00Z"
+        sourceRef: "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-amazon-nova-2-lite.html — per-model Input Modalities table marks Text, Image, Video supported; PDF is not listed as a separate modality",
+        confidence: "declared",
+        observedAt: "2026-09-25T12:31:56.381Z"
       },
       outputModalities: {
         value: [
@@ -25807,6 +26297,20 @@ var catalog_default = {
       },
       unsupportedParameters: [],
       status: "candidate",
+      promptCacheKey: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-25T19:30:00Z",
+        sourceRef: "https://github.com/openai/codex/blob/main/codex-rs/core/src/client.rs — Codex (the vendor's own client for this backend) sets prompt_cache_key on every Responses request, to the session id (or source:parent_thread_id for internal sessions); https://developers.openai.com/api/docs/guides/prompt-caching documents the field."
+      },
+      clientToolSearch: {
+        value: true,
+        source: "upstream-static",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: "codex-rs models-manager/models.json (cloned 25270df, 2026-09-26): supports_search_tool is true on every bundled model; core/src/tools/spec_plan.rs:653-666 turns on the client tool_search (tools/src/tool_search.rs:84-110, core/src/tools/handlers/tool_search_spec.rs:98) when the provider's namespace capability is on (default true) -- the vendor's own client sends it to chatgpt.com/backend-api/codex."
+      },
       canonicalModelId: "gpt-5.6-luna",
       modelFamily: "gpt"
     },
@@ -25914,6 +26418,20 @@ var catalog_default = {
       },
       unsupportedParameters: [],
       status: "candidate",
+      promptCacheKey: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-25T19:30:00Z",
+        sourceRef: "https://github.com/openai/codex/blob/main/codex-rs/core/src/client.rs — Codex (the vendor's own client for this backend) sets prompt_cache_key on every Responses request, to the session id (or source:parent_thread_id for internal sessions); https://developers.openai.com/api/docs/guides/prompt-caching documents the field."
+      },
+      clientToolSearch: {
+        value: true,
+        source: "upstream-static",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: "codex-rs models-manager/models.json (cloned 25270df, 2026-09-26): supports_search_tool is true on every bundled model; core/src/tools/spec_plan.rs:653-666 turns on the client tool_search (tools/src/tool_search.rs:84-110, core/src/tools/handlers/tool_search_spec.rs:98) when the provider's namespace capability is on (default true) -- the vendor's own client sends it to chatgpt.com/backend-api/codex."
+      },
       canonicalModelId: "gpt-5.6-sol",
       modelFamily: "gpt"
     },
@@ -26021,6 +26539,20 @@ var catalog_default = {
       },
       unsupportedParameters: [],
       status: "candidate",
+      promptCacheKey: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-25T19:30:00Z",
+        sourceRef: "https://github.com/openai/codex/blob/main/codex-rs/core/src/client.rs — Codex (the vendor's own client for this backend) sets prompt_cache_key on every Responses request, to the session id (or source:parent_thread_id for internal sessions); https://developers.openai.com/api/docs/guides/prompt-caching documents the field."
+      },
+      clientToolSearch: {
+        value: true,
+        source: "upstream-static",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: "codex-rs models-manager/models.json (cloned 25270df, 2026-09-26): supports_search_tool is true on every bundled model; core/src/tools/spec_plan.rs:653-666 turns on the client tool_search (tools/src/tool_search.rs:84-110, core/src/tools/handlers/tool_search_spec.rs:98) when the provider's namespace capability is on (default true) -- the vendor's own client sends it to chatgpt.com/backend-api/codex."
+      },
       canonicalModelId: "gpt-5.6-terra",
       modelFamily: "gpt"
     },
@@ -26130,6 +26662,20 @@ var catalog_default = {
       },
       unsupportedParameters: [],
       status: "candidate",
+      promptCacheKey: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-25T19:30:00Z",
+        sourceRef: "https://github.com/openai/codex/blob/main/codex-rs/core/src/client.rs — Codex (the vendor's own client for this backend) sets prompt_cache_key on every Responses request, to the session id (or source:parent_thread_id for internal sessions); https://developers.openai.com/api/docs/guides/prompt-caching documents the field."
+      },
+      clientToolSearch: {
+        value: true,
+        source: "upstream-static",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: "codex-rs models-manager/models.json (cloned 25270df, 2026-09-26): supports_search_tool is true on every bundled model; core/src/tools/spec_plan.rs:653-666 turns on the client tool_search (tools/src/tool_search.rs:84-110, core/src/tools/handlers/tool_search_spec.rs:98) when the provider's namespace capability is on (default true) -- the vendor's own client sends it to chatgpt.com/backend-api/codex."
+      },
       canonicalModelId: "gpt-6-astra",
       modelFamily: "gpt"
     },
@@ -26239,6 +26785,20 @@ var catalog_default = {
       },
       unsupportedParameters: [],
       status: "candidate",
+      promptCacheKey: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-25T19:30:00Z",
+        sourceRef: "https://github.com/openai/codex/blob/main/codex-rs/core/src/client.rs — Codex (the vendor's own client for this backend) sets prompt_cache_key on every Responses request, to the session id (or source:parent_thread_id for internal sessions); https://developers.openai.com/api/docs/guides/prompt-caching documents the field."
+      },
+      clientToolSearch: {
+        value: true,
+        source: "upstream-static",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: "codex-rs models-manager/models.json (cloned 25270df, 2026-09-26): supports_search_tool is true on every bundled model; core/src/tools/spec_plan.rs:653-666 turns on the client tool_search (tools/src/tool_search.rs:84-110, core/src/tools/handlers/tool_search_spec.rs:98) when the provider's namespace capability is on (default true) -- the vendor's own client sends it to chatgpt.com/backend-api/codex."
+      },
       canonicalModelId: "gpt-6-luna",
       modelFamily: "gpt"
     },
@@ -26348,6 +26908,20 @@ var catalog_default = {
       },
       unsupportedParameters: [],
       status: "candidate",
+      promptCacheKey: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-25T19:30:00Z",
+        sourceRef: "https://github.com/openai/codex/blob/main/codex-rs/core/src/client.rs — Codex (the vendor's own client for this backend) sets prompt_cache_key on every Responses request, to the session id (or source:parent_thread_id for internal sessions); https://developers.openai.com/api/docs/guides/prompt-caching documents the field."
+      },
+      clientToolSearch: {
+        value: true,
+        source: "upstream-static",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: "codex-rs models-manager/models.json (cloned 25270df, 2026-09-26): supports_search_tool is true on every bundled model; core/src/tools/spec_plan.rs:653-666 turns on the client tool_search (tools/src/tool_search.rs:84-110, core/src/tools/handlers/tool_search_spec.rs:98) when the provider's namespace capability is on (default true) -- the vendor's own client sends it to chatgpt.com/backend-api/codex."
+      },
       canonicalModelId: "gpt-6-sol",
       modelFamily: "gpt"
     },
@@ -27060,7 +27634,16 @@ var catalog_default = {
           "max"
         ],
         continuation: "opaque-provider-state",
-        defaultEffort: "high"
+        defaultEffort: "high",
+        effortRequest: {
+          value: {
+            field: "output_config.effort"
+          },
+          source: "official-doc",
+          confidence: "declared",
+          observedAt: "2026-09-25T12:30:00Z",
+          sourceRef: "https://platform.claude.com/docs/en/build-with-claude/effort — the effort page's supportedModels list names claude-fable-5-1, claude-fable-5, claude-opus-5-5, claude-opus-5, claude-opus-4-8, claude-opus-4-7, claude-opus-4-6, claude-opus-4-5-20251101, claude-sonnet-5 and claude-sonnet-4-6: effort is requested through `output_config.effort` (Claude 4.7 and later reject a manual `thinking.budget_tokens`, https://platform.claude.com/docs/en/build-with-claude/thinking-troubleshooting#rejected-configurations)."
+        }
       },
       pricing: {
         value: {
@@ -27077,9 +27660,50 @@ var catalog_default = {
       unsupportedParameters: [
         "temperature",
         "top_p",
-        "top_k"
+        "top_k",
+        "thinking.type.enabled",
+        "thinking.type.disabled"
       ],
       status: "candidate",
+      deferredToolLoading: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-25T18:30:00Z",
+        sourceRef: 'https://platform.claude.com/docs/en/agents-and-tools/tool-use/tool-search-tool — the Model compatibility table lists Fable 5.1, Mythos 5.1, Fable 5, Mythos 5, Opus 5.5, Opus 5, Opus 4.8, 4.7, 4.6, Sonnet 4.6, Opus 4.5, Sonnet 4.5 and Haiku 4.5 (not Sonnet 5); "Custom tool search implementation": a client tool returns a tool_result whose content carries {type:"tool_reference", tool_name} blocks, every referenced tool declared in tools with defer_loading: true, and the API expands them; "The prefix is untouched, so prompt caching is preserved" (https://platform.claude.com/docs/en/agents-and-tools/tool-use/tool-use-with-prompt-caching).'
+      },
+      midConversationSystem: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-25T19:00:00Z",
+        sourceRef: 'https://platform.claude.com/docs/en/build-with-claude/mid-conversation-system-messages — "This feature is available on Claude Fable 5.1, Claude Mythos 5.1, Claude Fable 5, Claude Mythos 5, Claude Opus 5.5, Claude Opus 4.8, and Claude Opus 5. No beta header is required for mid-conversation system messages. This feature is not available on Claude Sonnet 5."'
+      },
+      midConversationToolChanges: {
+        value: {
+          beta: "mid-conversation-tool-changes-2026-07-01"
+        },
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: 'https://platform.claude.com/docs/en/build-with-claude/mid-conversation-system-messages — tool_addition / tool_removal blocks in a role:"system" message, by reference ("Add/remove by reference: mid-conversation-tool-changes-2026-07-01 — Claude API, Amazon Bedrock, Google Cloud"); available on Claude Fable 5.1, Mythos 5.1, Fable 5, Mythos 5, Opus 5.5, Opus 4.8 and Opus 5, "Not available on Claude Sonnet 5". Retrieved 2026-09-26; release note 2026-07-24 (https://platform.claude.com/docs/en/release-notes/overview).'
+      },
+      inlineToolDefinitions: {
+        value: {
+          beta: "inline-tools-2026-09-15"
+        },
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: 'https://platform.claude.com/docs/en/build-with-claude/mid-conversation-system-messages — tool_addition with a tool_definition ("Add/remove by reference + define by value: inline-tools-2026-09-15 — Claude API"); "The inline-tools-2026-09-15 header covers all reference-based changes"; redefinition: "send a different definition under the same name ... The new definition replaces the earlier one from that position onward" (no removal first, a tool declared in tools included); a definition reusing the name of a different type of tool is a 400 tool_name_conflict. Same model list as midConversationToolChanges. Claude API only, so set on the anthropic and console providers (both api.anthropic.com), never on a cloud reseller. Retrieved 2026-09-26; release note 2026-09-22.'
+      },
+      assistantPrefill: {
+        value: false,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: `https://platform.claude.com/docs/en/models/opus-5-5/migration-guide — "Don't end messages with a prefilled assistant turn: it is rejected"; "Prefilling assistant messages returns a 400 error on Claude Opus 4.6 and later Opus models, including Claude Opus 5.5"; Sonnet 5 and Opus 5: "Manual extended thinking, non-default sampling parameters, and assistant prefill return a 400 error on both models"; the Fable 5.1 guide: prefill "returns a 400 error, unchanged from Claude Fable 5". Live gate 2026-09-26 on claude-opus-5-5: "This model does not support assistant message prefill. The conversation must end with a user message."`
+      },
       canonicalModelId: "claude-fable-5",
       modelFamily: "claude"
     },
@@ -27212,6 +27836,33 @@ var catalog_default = {
           sourceRef: `continuity report §4.4, BOTH halves, applied per anthropic/claude-opus-5's own evidence. Own-state acceptance: a Claude model's own thinking blocks are replayed to it unchanged, in order, with signatures intact. Why the domain is NARROW: prior thinking/redacted_thinking blocks are tied to the model that produced them, so "same provider" is not automatically "same continuation domain" — the domain is this model alone.`,
           confidence: "declared",
           observedAt: "2026-09-16T00:00:00Z"
+        },
+        effortRequest: {
+          value: {
+            field: "output_config.effort"
+          },
+          source: "official-doc",
+          confidence: "declared",
+          observedAt: "2026-09-25T12:30:00Z",
+          sourceRef: "https://platform.claude.com/docs/en/build-with-claude/effort — the effort page's supportedModels list names claude-fable-5-1, claude-fable-5, claude-opus-5-5, claude-opus-5, claude-opus-4-8, claude-opus-4-7, claude-opus-4-6, claude-opus-4-5-20251101, claude-sonnet-5 and claude-sonnet-4-6: effort is requested through `output_config.effort` (Claude 4.7 and later reject a manual `thinking.budget_tokens`, https://platform.claude.com/docs/en/build-with-claude/thinking-troubleshooting#rejected-configurations)."
+        },
+        blockBinding: {
+          value: {
+            beta: "thinking-binding-controls-2026-08-01"
+          },
+          source: "official-doc",
+          confidence: "declared",
+          observedAt: "2026-09-25T13:00:00Z",
+          sourceRef: 'https://platform.claude.com/docs/en/build-with-claude/thinking-troubleshooting — "A 400 error says a thinking block signature is invalid": on Claude Fable 5.1 and Opus 5.5 a replayed thinking block is bound to the conversation (system, tools and earlier messages) and is rejected once that prefix changes (enforced for accounts created on or after 2026-08-31); the documented escape is the `thinking-binding-controls-2026-08-01` beta with `thinking.block_binding.prefix_mismatch_behavior: "drop_block"`. See also https://platform.claude.com/docs/en/models/opus-5-5/whats-new-opus-5-5 ("Thinking blocks are tied to the model and the conversation").'
+        },
+        perMessageEffort: {
+          value: {
+            beta: "mid-conversation-output-config-2026-07-01"
+          },
+          source: "official-doc",
+          confidence: "declared",
+          observedAt: "2026-09-25T18:00:00Z",
+          sourceRef: 'https://platform.claude.com/docs/en/build-with-claude/effort#change-effort-mid-conversation-beta — "On Claude Fable 5.1, Claude Mythos 5.1, Claude Opus 5.5, and Claude Opus 5, use a per-message effort change, which keeps the prompt cache": a role:"system" message with empty content and output_config.effort, beta header mid-conversation-output-config-2026-07-01; "Models without per-message effort, including Claude Fable 5, return a 400 error". claude 2.1.282 sends the older alias per-turn-control-2026-07-01, which the page does not document.'
         }
       },
       pricing: {
@@ -27226,8 +27877,52 @@ var catalog_default = {
         confidence: "declared",
         observedAt: "2026-09-16T00:00:00Z"
       },
-      unsupportedParameters: [],
+      unsupportedParameters: [
+        "thinking.type.enabled",
+        "thinking.type.disabled",
+        "tool_choice.any",
+        "tool_choice.tool"
+      ],
       status: "candidate",
+      deferredToolLoading: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-25T18:30:00Z",
+        sourceRef: 'https://platform.claude.com/docs/en/agents-and-tools/tool-use/tool-search-tool — the Model compatibility table lists Fable 5.1, Mythos 5.1, Fable 5, Mythos 5, Opus 5.5, Opus 5, Opus 4.8, 4.7, 4.6, Sonnet 4.6, Opus 4.5, Sonnet 4.5 and Haiku 4.5 (not Sonnet 5); "Custom tool search implementation": a client tool returns a tool_result whose content carries {type:"tool_reference", tool_name} blocks, every referenced tool declared in tools with defer_loading: true, and the API expands them; "The prefix is untouched, so prompt caching is preserved" (https://platform.claude.com/docs/en/agents-and-tools/tool-use/tool-use-with-prompt-caching).'
+      },
+      midConversationSystem: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-25T19:00:00Z",
+        sourceRef: 'https://platform.claude.com/docs/en/build-with-claude/mid-conversation-system-messages — "This feature is available on Claude Fable 5.1, Claude Mythos 5.1, Claude Fable 5, Claude Mythos 5, Claude Opus 5.5, Claude Opus 4.8, and Claude Opus 5. No beta header is required for mid-conversation system messages. This feature is not available on Claude Sonnet 5."'
+      },
+      midConversationToolChanges: {
+        value: {
+          beta: "mid-conversation-tool-changes-2026-07-01"
+        },
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: 'https://platform.claude.com/docs/en/build-with-claude/mid-conversation-system-messages — tool_addition / tool_removal blocks in a role:"system" message, by reference ("Add/remove by reference: mid-conversation-tool-changes-2026-07-01 — Claude API, Amazon Bedrock, Google Cloud"); available on Claude Fable 5.1, Mythos 5.1, Fable 5, Mythos 5, Opus 5.5, Opus 4.8 and Opus 5, "Not available on Claude Sonnet 5". Retrieved 2026-09-26; release note 2026-07-24 (https://platform.claude.com/docs/en/release-notes/overview).'
+      },
+      inlineToolDefinitions: {
+        value: {
+          beta: "inline-tools-2026-09-15"
+        },
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: 'https://platform.claude.com/docs/en/build-with-claude/mid-conversation-system-messages — tool_addition with a tool_definition ("Add/remove by reference + define by value: inline-tools-2026-09-15 — Claude API"); "The inline-tools-2026-09-15 header covers all reference-based changes"; redefinition: "send a different definition under the same name ... The new definition replaces the earlier one from that position onward" (no removal first, a tool declared in tools included); a definition reusing the name of a different type of tool is a 400 tool_name_conflict. Same model list as midConversationToolChanges. Claude API only, so set on the anthropic and console providers (both api.anthropic.com), never on a cloud reseller. Retrieved 2026-09-26; release note 2026-09-22.'
+      },
+      assistantPrefill: {
+        value: false,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: `https://platform.claude.com/docs/en/models/opus-5-5/migration-guide — "Don't end messages with a prefilled assistant turn: it is rejected"; "Prefilling assistant messages returns a 400 error on Claude Opus 4.6 and later Opus models, including Claude Opus 5.5"; Sonnet 5 and Opus 5: "Manual extended thinking, non-default sampling parameters, and assistant prefill return a 400 error on both models"; the Fable 5.1 guide: prefill "returns a 400 error, unchanged from Claude Fable 5". Live gate 2026-09-26 on claude-opus-5-5: "This model does not support assistant message prefill. The conversation must end with a user message."`
+      },
       canonicalModelId: "claude-fable-5.1",
       modelFamily: "claude"
     },
@@ -27322,8 +28017,17 @@ var catalog_default = {
         confidence: "declared",
         observedAt: "2026-09-16T00:00:00Z"
       },
-      unsupportedParameters: [],
+      unsupportedParameters: [
+        "thinking.type.adaptive"
+      ],
       status: "candidate",
+      deferredToolLoading: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-25T18:30:00Z",
+        sourceRef: 'https://platform.claude.com/docs/en/agents-and-tools/tool-use/tool-search-tool — the Model compatibility table lists Fable 5.1, Mythos 5.1, Fable 5, Mythos 5, Opus 5.5, Opus 5, Opus 4.8, 4.7, 4.6, Sonnet 4.6, Opus 4.5, Sonnet 4.5 and Haiku 4.5 (not Sonnet 5); "Custom tool search implementation": a client tool returns a tool_result whose content carries {type:"tool_reference", tool_name} blocks, every referenced tool declared in tools with defer_loading: true, and the API expands them; "The prefix is untouched, so prompt caching is preserved" (https://platform.claude.com/docs/en/agents-and-tools/tool-use/tool-use-with-prompt-caching).'
+      },
       canonicalModelId: "claude-haiku-4.5-20251001",
       modelFamily: "claude"
     },
@@ -27418,8 +28122,17 @@ var catalog_default = {
         confidence: "declared",
         observedAt: "2026-09-19T00:00:00Z"
       },
-      unsupportedParameters: [],
+      unsupportedParameters: [
+        "thinking.type.adaptive"
+      ],
       status: "candidate",
+      deferredToolLoading: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-25T18:30:00Z",
+        sourceRef: 'https://platform.claude.com/docs/en/agents-and-tools/tool-use/tool-search-tool — the Model compatibility table lists Fable 5.1, Mythos 5.1, Fable 5, Mythos 5, Opus 5.5, Opus 5, Opus 4.8, 4.7, 4.6, Sonnet 4.6, Opus 4.5, Sonnet 4.5 and Haiku 4.5 (not Sonnet 5); "Custom tool search implementation": a client tool returns a tool_result whose content carries {type:"tool_reference", tool_name} blocks, every referenced tool declared in tools with defer_loading: true, and the API expands them; "The prefix is untouched, so prompt caching is preserved" (https://platform.claude.com/docs/en/agents-and-tools/tool-use/tool-use-with-prompt-caching).'
+      },
       canonicalModelId: "claude-haiku-4.5",
       modelFamily: "claude"
     },
@@ -27517,7 +28230,16 @@ var catalog_default = {
           "high"
         ],
         continuation: "opaque-provider-state",
-        defaultEffort: "high"
+        defaultEffort: "high",
+        effortRequest: {
+          value: {
+            field: "output_config.effort"
+          },
+          source: "official-doc",
+          confidence: "declared",
+          observedAt: "2026-09-25T12:30:00Z",
+          sourceRef: "https://platform.claude.com/docs/en/build-with-claude/effort — the effort page's supportedModels list names claude-fable-5-1, claude-fable-5, claude-opus-5-5, claude-opus-5, claude-opus-4-8, claude-opus-4-7, claude-opus-4-6, claude-opus-4-5-20251101, claude-sonnet-5 and claude-sonnet-4-6: effort is requested through `output_config.effort` (Claude 4.7 and later reject a manual `thinking.budget_tokens`, https://platform.claude.com/docs/en/build-with-claude/thinking-troubleshooting#rejected-configurations)."
+        }
       },
       pricing: {
         value: {
@@ -27531,8 +28253,17 @@ var catalog_default = {
         confidence: "declared",
         observedAt: "2026-09-19T00:00:00Z"
       },
-      unsupportedParameters: [],
+      unsupportedParameters: [
+        "thinking.type.adaptive"
+      ],
       status: "candidate",
+      deferredToolLoading: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-25T18:30:00Z",
+        sourceRef: 'https://platform.claude.com/docs/en/agents-and-tools/tool-use/tool-search-tool — the Model compatibility table lists Fable 5.1, Mythos 5.1, Fable 5, Mythos 5, Opus 5.5, Opus 5, Opus 4.8, 4.7, 4.6, Sonnet 4.6, Opus 4.5, Sonnet 4.5 and Haiku 4.5 (not Sonnet 5); "Custom tool search implementation": a client tool returns a tool_result whose content carries {type:"tool_reference", tool_name} blocks, every referenced tool declared in tools with defer_loading: true, and the API expands them; "The prefix is untouched, so prompt caching is preserved" (https://platform.claude.com/docs/en/agents-and-tools/tool-use/tool-use-with-prompt-caching).'
+      },
       canonicalModelId: "claude-opus-4.5",
       modelFamily: "claude"
     },
@@ -27541,7 +28272,9 @@ var catalog_default = {
       providerId: "console",
       upstreamId: "claude-opus-4.6",
       displayName: "Claude Opus 4.6",
-      aliases: [],
+      aliases: [
+        "claude-opus-4-6"
+      ],
       endpoints: [
         "chat"
       ],
@@ -27628,7 +28361,16 @@ var catalog_default = {
           "max"
         ],
         continuation: "opaque-provider-state",
-        defaultEffort: "high"
+        defaultEffort: "high",
+        effortRequest: {
+          value: {
+            field: "output_config.effort"
+          },
+          source: "official-doc",
+          confidence: "declared",
+          observedAt: "2026-09-25T12:30:00Z",
+          sourceRef: "https://platform.claude.com/docs/en/build-with-claude/effort — the effort page's supportedModels list names claude-fable-5-1, claude-fable-5, claude-opus-5-5, claude-opus-5, claude-opus-4-8, claude-opus-4-7, claude-opus-4-6, claude-opus-4-5-20251101, claude-sonnet-5 and claude-sonnet-4-6: effort is requested through `output_config.effort` (Claude 4.7 and later reject a manual `thinking.budget_tokens`, https://platform.claude.com/docs/en/build-with-claude/thinking-troubleshooting#rejected-configurations)."
+        }
       },
       pricing: {
         value: {
@@ -27644,6 +28386,20 @@ var catalog_default = {
       },
       unsupportedParameters: [],
       status: "candidate",
+      deferredToolLoading: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-25T18:30:00Z",
+        sourceRef: 'https://platform.claude.com/docs/en/agents-and-tools/tool-use/tool-search-tool — the Model compatibility table lists Fable 5.1, Mythos 5.1, Fable 5, Mythos 5, Opus 5.5, Opus 5, Opus 4.8, 4.7, 4.6, Sonnet 4.6, Opus 4.5, Sonnet 4.5 and Haiku 4.5 (not Sonnet 5); "Custom tool search implementation": a client tool returns a tool_result whose content carries {type:"tool_reference", tool_name} blocks, every referenced tool declared in tools with defer_loading: true, and the API expands them; "The prefix is untouched, so prompt caching is preserved" (https://platform.claude.com/docs/en/agents-and-tools/tool-use/tool-use-with-prompt-caching).'
+      },
+      assistantPrefill: {
+        value: false,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: `https://platform.claude.com/docs/en/models/opus-5-5/migration-guide — "Don't end messages with a prefilled assistant turn: it is rejected"; "Prefilling assistant messages returns a 400 error on Claude Opus 4.6 and later Opus models, including Claude Opus 5.5"; Sonnet 5 and Opus 5: "Manual extended thinking, non-default sampling parameters, and assistant prefill return a 400 error on both models"; the Fable 5.1 guide: prefill "returns a 400 error, unchanged from Claude Fable 5". Live gate 2026-09-26 on claude-opus-5-5: "This model does not support assistant message prefill. The conversation must end with a user message."`
+      },
       canonicalModelId: "claude-opus-4.6",
       modelFamily: "claude"
     },
@@ -27652,7 +28408,9 @@ var catalog_default = {
       providerId: "console",
       upstreamId: "claude-opus-4.7",
       displayName: "Claude Opus 4.7",
-      aliases: [],
+      aliases: [
+        "claude-opus-4-7"
+      ],
       endpoints: [
         "chat"
       ],
@@ -27740,7 +28498,16 @@ var catalog_default = {
           "max"
         ],
         continuation: "opaque-provider-state",
-        defaultEffort: "high"
+        defaultEffort: "high",
+        effortRequest: {
+          value: {
+            field: "output_config.effort"
+          },
+          source: "official-doc",
+          confidence: "declared",
+          observedAt: "2026-09-25T12:30:00Z",
+          sourceRef: "https://platform.claude.com/docs/en/build-with-claude/effort — the effort page's supportedModels list names claude-fable-5-1, claude-fable-5, claude-opus-5-5, claude-opus-5, claude-opus-4-8, claude-opus-4-7, claude-opus-4-6, claude-opus-4-5-20251101, claude-sonnet-5 and claude-sonnet-4-6: effort is requested through `output_config.effort` (Claude 4.7 and later reject a manual `thinking.budget_tokens`, https://platform.claude.com/docs/en/build-with-claude/thinking-troubleshooting#rejected-configurations)."
+        }
       },
       pricing: {
         value: {
@@ -27757,9 +28524,24 @@ var catalog_default = {
       unsupportedParameters: [
         "temperature",
         "top_p",
-        "top_k"
+        "top_k",
+        "thinking.type.enabled"
       ],
       status: "candidate",
+      deferredToolLoading: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-25T18:30:00Z",
+        sourceRef: 'https://platform.claude.com/docs/en/agents-and-tools/tool-use/tool-search-tool — the Model compatibility table lists Fable 5.1, Mythos 5.1, Fable 5, Mythos 5, Opus 5.5, Opus 5, Opus 4.8, 4.7, 4.6, Sonnet 4.6, Opus 4.5, Sonnet 4.5 and Haiku 4.5 (not Sonnet 5); "Custom tool search implementation": a client tool returns a tool_result whose content carries {type:"tool_reference", tool_name} blocks, every referenced tool declared in tools with defer_loading: true, and the API expands them; "The prefix is untouched, so prompt caching is preserved" (https://platform.claude.com/docs/en/agents-and-tools/tool-use/tool-use-with-prompt-caching).'
+      },
+      assistantPrefill: {
+        value: false,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: `https://platform.claude.com/docs/en/models/opus-5-5/migration-guide — "Don't end messages with a prefilled assistant turn: it is rejected"; "Prefilling assistant messages returns a 400 error on Claude Opus 4.6 and later Opus models, including Claude Opus 5.5"; Sonnet 5 and Opus 5: "Manual extended thinking, non-default sampling parameters, and assistant prefill return a 400 error on both models"; the Fable 5.1 guide: prefill "returns a 400 error, unchanged from Claude Fable 5". Live gate 2026-09-26 on claude-opus-5-5: "This model does not support assistant message prefill. The conversation must end with a user message."`
+      },
       canonicalModelId: "claude-opus-4.7",
       modelFamily: "claude"
     },
@@ -27768,7 +28550,9 @@ var catalog_default = {
       providerId: "console",
       upstreamId: "claude-opus-4.8",
       displayName: "Claude Opus 4.8",
-      aliases: [],
+      aliases: [
+        "claude-opus-4-8"
+      ],
       endpoints: [
         "chat"
       ],
@@ -27856,7 +28640,16 @@ var catalog_default = {
           "max"
         ],
         continuation: "opaque-provider-state",
-        defaultEffort: "high"
+        defaultEffort: "high",
+        effortRequest: {
+          value: {
+            field: "output_config.effort"
+          },
+          source: "official-doc",
+          confidence: "declared",
+          observedAt: "2026-09-25T12:30:00Z",
+          sourceRef: "https://platform.claude.com/docs/en/build-with-claude/effort — the effort page's supportedModels list names claude-fable-5-1, claude-fable-5, claude-opus-5-5, claude-opus-5, claude-opus-4-8, claude-opus-4-7, claude-opus-4-6, claude-opus-4-5-20251101, claude-sonnet-5 and claude-sonnet-4-6: effort is requested through `output_config.effort` (Claude 4.7 and later reject a manual `thinking.budget_tokens`, https://platform.claude.com/docs/en/build-with-claude/thinking-troubleshooting#rejected-configurations)."
+        }
       },
       pricing: {
         value: {
@@ -27873,9 +28666,49 @@ var catalog_default = {
       unsupportedParameters: [
         "temperature",
         "top_p",
-        "top_k"
+        "top_k",
+        "thinking.type.enabled"
       ],
       status: "candidate",
+      deferredToolLoading: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-25T18:30:00Z",
+        sourceRef: 'https://platform.claude.com/docs/en/agents-and-tools/tool-use/tool-search-tool — the Model compatibility table lists Fable 5.1, Mythos 5.1, Fable 5, Mythos 5, Opus 5.5, Opus 5, Opus 4.8, 4.7, 4.6, Sonnet 4.6, Opus 4.5, Sonnet 4.5 and Haiku 4.5 (not Sonnet 5); "Custom tool search implementation": a client tool returns a tool_result whose content carries {type:"tool_reference", tool_name} blocks, every referenced tool declared in tools with defer_loading: true, and the API expands them; "The prefix is untouched, so prompt caching is preserved" (https://platform.claude.com/docs/en/agents-and-tools/tool-use/tool-use-with-prompt-caching).'
+      },
+      midConversationSystem: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-25T19:00:00Z",
+        sourceRef: 'https://platform.claude.com/docs/en/build-with-claude/mid-conversation-system-messages — "This feature is available on Claude Fable 5.1, Claude Mythos 5.1, Claude Fable 5, Claude Mythos 5, Claude Opus 5.5, Claude Opus 4.8, and Claude Opus 5. No beta header is required for mid-conversation system messages. This feature is not available on Claude Sonnet 5."'
+      },
+      midConversationToolChanges: {
+        value: {
+          beta: "mid-conversation-tool-changes-2026-07-01"
+        },
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: 'https://platform.claude.com/docs/en/build-with-claude/mid-conversation-system-messages — tool_addition / tool_removal blocks in a role:"system" message, by reference ("Add/remove by reference: mid-conversation-tool-changes-2026-07-01 — Claude API, Amazon Bedrock, Google Cloud"); available on Claude Fable 5.1, Mythos 5.1, Fable 5, Mythos 5, Opus 5.5, Opus 4.8 and Opus 5, "Not available on Claude Sonnet 5". Retrieved 2026-09-26; release note 2026-07-24 (https://platform.claude.com/docs/en/release-notes/overview).'
+      },
+      inlineToolDefinitions: {
+        value: {
+          beta: "inline-tools-2026-09-15"
+        },
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: 'https://platform.claude.com/docs/en/build-with-claude/mid-conversation-system-messages — tool_addition with a tool_definition ("Add/remove by reference + define by value: inline-tools-2026-09-15 — Claude API"); "The inline-tools-2026-09-15 header covers all reference-based changes"; redefinition: "send a different definition under the same name ... The new definition replaces the earlier one from that position onward" (no removal first, a tool declared in tools included); a definition reusing the name of a different type of tool is a 400 tool_name_conflict. Same model list as midConversationToolChanges. Claude API only, so set on the anthropic and console providers (both api.anthropic.com), never on a cloud reseller. Retrieved 2026-09-26; release note 2026-09-22.'
+      },
+      assistantPrefill: {
+        value: false,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: `https://platform.claude.com/docs/en/models/opus-5-5/migration-guide — "Don't end messages with a prefilled assistant turn: it is rejected"; "Prefilling assistant messages returns a 400 error on Claude Opus 4.6 and later Opus models, including Claude Opus 5.5"; Sonnet 5 and Opus 5: "Manual extended thinking, non-default sampling parameters, and assistant prefill return a 400 error on both models"; the Fable 5.1 guide: prefill "returns a 400 error, unchanged from Claude Fable 5". Live gate 2026-09-26 on claude-opus-5-5: "This model does not support assistant message prefill. The conversation must end with a user message."`
+      },
       canonicalModelId: "claude-opus-4.8",
       modelFamily: "claude"
     },
@@ -28015,6 +28848,24 @@ var catalog_default = {
           sourceRef: "continuity report §4.4, BOTH halves. Own-state acceptance: the model's own `thinking` blocks are replayed to it unchanged, in order, with their signatures intact (§4.4's replay rules and its worked request). Why the domain is NARROW: \"when changing Claude models, prior `thinking` and `redacted_thinking` blocks should be stripped because they are tied to the model that produced them. Therefore 'same provider' is not automatically 'same continuation domain.'\" The domain is this model alone, never the Anthropic provider.",
           confidence: "declared",
           observedAt: "2026-09-16T00:00:00Z"
+        },
+        effortRequest: {
+          value: {
+            field: "output_config.effort"
+          },
+          source: "official-doc",
+          confidence: "declared",
+          observedAt: "2026-09-25T12:30:00Z",
+          sourceRef: "https://platform.claude.com/docs/en/build-with-claude/effort — the effort page's supportedModels list names claude-fable-5-1, claude-fable-5, claude-opus-5-5, claude-opus-5, claude-opus-4-8, claude-opus-4-7, claude-opus-4-6, claude-opus-4-5-20251101, claude-sonnet-5 and claude-sonnet-4-6: effort is requested through `output_config.effort` (Claude 4.7 and later reject a manual `thinking.budget_tokens`, https://platform.claude.com/docs/en/build-with-claude/thinking-troubleshooting#rejected-configurations)."
+        },
+        perMessageEffort: {
+          value: {
+            beta: "mid-conversation-output-config-2026-07-01"
+          },
+          source: "official-doc",
+          confidence: "declared",
+          observedAt: "2026-09-25T18:00:00Z",
+          sourceRef: 'https://platform.claude.com/docs/en/build-with-claude/effort#change-effort-mid-conversation-beta — "On Claude Fable 5.1, Claude Mythos 5.1, Claude Opus 5.5, and Claude Opus 5, use a per-message effort change, which keeps the prompt cache": a role:"system" message with empty content and output_config.effort, beta header mid-conversation-output-config-2026-07-01; "Models without per-message effort, including Claude Fable 5, return a 400 error". claude 2.1.282 sends the older alias per-turn-control-2026-07-01, which the page does not document.'
         }
       },
       pricing: {
@@ -28032,9 +28883,51 @@ var catalog_default = {
       unsupportedParameters: [
         "temperature",
         "top_p",
-        "top_k"
+        "top_k",
+        "thinking.type.enabled",
+        "thinking.type.disabled+output_config.effort.xhigh",
+        "thinking.type.disabled+output_config.effort.max"
       ],
       status: "candidate",
+      deferredToolLoading: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-25T18:30:00Z",
+        sourceRef: 'https://platform.claude.com/docs/en/agents-and-tools/tool-use/tool-search-tool — the Model compatibility table lists Fable 5.1, Mythos 5.1, Fable 5, Mythos 5, Opus 5.5, Opus 5, Opus 4.8, 4.7, 4.6, Sonnet 4.6, Opus 4.5, Sonnet 4.5 and Haiku 4.5 (not Sonnet 5); "Custom tool search implementation": a client tool returns a tool_result whose content carries {type:"tool_reference", tool_name} blocks, every referenced tool declared in tools with defer_loading: true, and the API expands them; "The prefix is untouched, so prompt caching is preserved" (https://platform.claude.com/docs/en/agents-and-tools/tool-use/tool-use-with-prompt-caching).'
+      },
+      midConversationSystem: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-25T19:00:00Z",
+        sourceRef: 'https://platform.claude.com/docs/en/build-with-claude/mid-conversation-system-messages — "This feature is available on Claude Fable 5.1, Claude Mythos 5.1, Claude Fable 5, Claude Mythos 5, Claude Opus 5.5, Claude Opus 4.8, and Claude Opus 5. No beta header is required for mid-conversation system messages. This feature is not available on Claude Sonnet 5."'
+      },
+      midConversationToolChanges: {
+        value: {
+          beta: "mid-conversation-tool-changes-2026-07-01"
+        },
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: 'https://platform.claude.com/docs/en/build-with-claude/mid-conversation-system-messages — tool_addition / tool_removal blocks in a role:"system" message, by reference ("Add/remove by reference: mid-conversation-tool-changes-2026-07-01 — Claude API, Amazon Bedrock, Google Cloud"); available on Claude Fable 5.1, Mythos 5.1, Fable 5, Mythos 5, Opus 5.5, Opus 4.8 and Opus 5, "Not available on Claude Sonnet 5". Retrieved 2026-09-26; release note 2026-07-24 (https://platform.claude.com/docs/en/release-notes/overview).'
+      },
+      inlineToolDefinitions: {
+        value: {
+          beta: "inline-tools-2026-09-15"
+        },
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: 'https://platform.claude.com/docs/en/build-with-claude/mid-conversation-system-messages — tool_addition with a tool_definition ("Add/remove by reference + define by value: inline-tools-2026-09-15 — Claude API"); "The inline-tools-2026-09-15 header covers all reference-based changes"; redefinition: "send a different definition under the same name ... The new definition replaces the earlier one from that position onward" (no removal first, a tool declared in tools included); a definition reusing the name of a different type of tool is a 400 tool_name_conflict. Same model list as midConversationToolChanges. Claude API only, so set on the anthropic and console providers (both api.anthropic.com), never on a cloud reseller. Retrieved 2026-09-26; release note 2026-09-22.'
+      },
+      assistantPrefill: {
+        value: false,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: `https://platform.claude.com/docs/en/models/opus-5-5/migration-guide — "Don't end messages with a prefilled assistant turn: it is rejected"; "Prefilling assistant messages returns a 400 error on Claude Opus 4.6 and later Opus models, including Claude Opus 5.5"; Sonnet 5 and Opus 5: "Manual extended thinking, non-default sampling parameters, and assistant prefill return a 400 error on both models"; the Fable 5.1 guide: prefill "returns a 400 error, unchanged from Claude Fable 5". Live gate 2026-09-26 on claude-opus-5-5: "This model does not support assistant message prefill. The conversation must end with a user message."`
+      },
       canonicalModelId: "claude-opus-5",
       modelFamily: "claude"
     },
@@ -28176,6 +29069,33 @@ var catalog_default = {
           sourceRef: "continuity report §3 (wire probe of the pinned Claude Agent SDK, 0.3.250→0.3.258) — every uncompacted historical thinking/redacted block is replayed on every later request",
           confidence: "declared",
           observedAt: "2026-09-05T00:00:00Z"
+        },
+        effortRequest: {
+          value: {
+            field: "output_config.effort"
+          },
+          source: "official-doc",
+          confidence: "declared",
+          observedAt: "2026-09-25T12:30:00Z",
+          sourceRef: "https://platform.claude.com/docs/en/build-with-claude/effort — the effort page's supportedModels list names claude-fable-5-1, claude-fable-5, claude-opus-5-5, claude-opus-5, claude-opus-4-8, claude-opus-4-7, claude-opus-4-6, claude-opus-4-5-20251101, claude-sonnet-5 and claude-sonnet-4-6: effort is requested through `output_config.effort` (Claude 4.7 and later reject a manual `thinking.budget_tokens`, https://platform.claude.com/docs/en/build-with-claude/thinking-troubleshooting#rejected-configurations)."
+        },
+        blockBinding: {
+          value: {
+            beta: "thinking-binding-controls-2026-08-01"
+          },
+          source: "official-doc",
+          confidence: "declared",
+          observedAt: "2026-09-25T13:00:00Z",
+          sourceRef: 'https://platform.claude.com/docs/en/build-with-claude/thinking-troubleshooting — "A 400 error says a thinking block signature is invalid": on Claude Fable 5.1 and Opus 5.5 a replayed thinking block is bound to the conversation (system, tools and earlier messages) and is rejected once that prefix changes (enforced for accounts created on or after 2026-08-31); the documented escape is the `thinking-binding-controls-2026-08-01` beta with `thinking.block_binding.prefix_mismatch_behavior: "drop_block"`. See also https://platform.claude.com/docs/en/models/opus-5-5/whats-new-opus-5-5 ("Thinking blocks are tied to the model and the conversation").'
+        },
+        perMessageEffort: {
+          value: {
+            beta: "mid-conversation-output-config-2026-07-01"
+          },
+          source: "official-doc",
+          confidence: "declared",
+          observedAt: "2026-09-25T18:00:00Z",
+          sourceRef: 'https://platform.claude.com/docs/en/build-with-claude/effort#change-effort-mid-conversation-beta — "On Claude Fable 5.1, Claude Mythos 5.1, Claude Opus 5.5, and Claude Opus 5, use a per-message effort change, which keeps the prompt cache": a role:"system" message with empty content and output_config.effort, beta header mid-conversation-output-config-2026-07-01; "Models without per-message effort, including Claude Fable 5, return a 400 error". claude 2.1.282 sends the older alias per-turn-control-2026-07-01, which the page does not document.'
         }
       },
       pricing: {
@@ -28190,8 +29110,52 @@ var catalog_default = {
         confidence: "declared",
         observedAt: "2026-09-25T10:25:20Z"
       },
-      unsupportedParameters: [],
+      unsupportedParameters: [
+        "thinking.type.enabled",
+        "thinking.type.disabled",
+        "tool_choice.any",
+        "tool_choice.tool"
+      ],
       status: "candidate",
+      deferredToolLoading: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-25T18:30:00Z",
+        sourceRef: 'https://platform.claude.com/docs/en/agents-and-tools/tool-use/tool-search-tool — the Model compatibility table lists Fable 5.1, Mythos 5.1, Fable 5, Mythos 5, Opus 5.5, Opus 5, Opus 4.8, 4.7, 4.6, Sonnet 4.6, Opus 4.5, Sonnet 4.5 and Haiku 4.5 (not Sonnet 5); "Custom tool search implementation": a client tool returns a tool_result whose content carries {type:"tool_reference", tool_name} blocks, every referenced tool declared in tools with defer_loading: true, and the API expands them; "The prefix is untouched, so prompt caching is preserved" (https://platform.claude.com/docs/en/agents-and-tools/tool-use/tool-use-with-prompt-caching).'
+      },
+      midConversationSystem: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-25T19:00:00Z",
+        sourceRef: 'https://platform.claude.com/docs/en/build-with-claude/mid-conversation-system-messages — "This feature is available on Claude Fable 5.1, Claude Mythos 5.1, Claude Fable 5, Claude Mythos 5, Claude Opus 5.5, Claude Opus 4.8, and Claude Opus 5. No beta header is required for mid-conversation system messages. This feature is not available on Claude Sonnet 5."'
+      },
+      midConversationToolChanges: {
+        value: {
+          beta: "mid-conversation-tool-changes-2026-07-01"
+        },
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: 'https://platform.claude.com/docs/en/build-with-claude/mid-conversation-system-messages — tool_addition / tool_removal blocks in a role:"system" message, by reference ("Add/remove by reference: mid-conversation-tool-changes-2026-07-01 — Claude API, Amazon Bedrock, Google Cloud"); available on Claude Fable 5.1, Mythos 5.1, Fable 5, Mythos 5, Opus 5.5, Opus 4.8 and Opus 5, "Not available on Claude Sonnet 5". Retrieved 2026-09-26; release note 2026-07-24 (https://platform.claude.com/docs/en/release-notes/overview).'
+      },
+      inlineToolDefinitions: {
+        value: {
+          beta: "inline-tools-2026-09-15"
+        },
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: 'https://platform.claude.com/docs/en/build-with-claude/mid-conversation-system-messages — tool_addition with a tool_definition ("Add/remove by reference + define by value: inline-tools-2026-09-15 — Claude API"); "The inline-tools-2026-09-15 header covers all reference-based changes"; redefinition: "send a different definition under the same name ... The new definition replaces the earlier one from that position onward" (no removal first, a tool declared in tools included); a definition reusing the name of a different type of tool is a 400 tool_name_conflict. Same model list as midConversationToolChanges. Claude API only, so set on the anthropic and console providers (both api.anthropic.com), never on a cloud reseller. Retrieved 2026-09-26; release note 2026-09-22.'
+      },
+      assistantPrefill: {
+        value: false,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: `https://platform.claude.com/docs/en/models/opus-5-5/migration-guide — "Don't end messages with a prefilled assistant turn: it is rejected"; "Prefilling assistant messages returns a 400 error on Claude Opus 4.6 and later Opus models, including Claude Opus 5.5"; Sonnet 5 and Opus 5: "Manual extended thinking, non-default sampling parameters, and assistant prefill return a 400 error on both models"; the Fable 5.1 guide: prefill "returns a 400 error, unchanged from Claude Fable 5". Live gate 2026-09-26 on claude-opus-5-5: "This model does not support assistant message prefill. The conversation must end with a user message."`
+      },
       canonicalModelId: "claude-opus-5.5",
       modelFamily: "claude"
     },
@@ -28303,8 +29267,17 @@ var catalog_default = {
         confidence: "declared",
         observedAt: "2026-09-19T00:00:00Z"
       },
-      unsupportedParameters: [],
+      unsupportedParameters: [
+        "thinking.type.adaptive"
+      ],
       status: "candidate",
+      deferredToolLoading: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-25T18:30:00Z",
+        sourceRef: 'https://platform.claude.com/docs/en/agents-and-tools/tool-use/tool-search-tool — the Model compatibility table lists Fable 5.1, Mythos 5.1, Fable 5, Mythos 5, Opus 5.5, Opus 5, Opus 4.8, 4.7, 4.6, Sonnet 4.6, Opus 4.5, Sonnet 4.5 and Haiku 4.5 (not Sonnet 5); "Custom tool search implementation": a client tool returns a tool_result whose content carries {type:"tool_reference", tool_name} blocks, every referenced tool declared in tools with defer_loading: true, and the API expands them; "The prefix is untouched, so prompt caching is preserved" (https://platform.claude.com/docs/en/agents-and-tools/tool-use/tool-use-with-prompt-caching).'
+      },
       canonicalModelId: "claude-sonnet-4.5",
       modelFamily: "claude"
     },
@@ -28313,7 +29286,9 @@ var catalog_default = {
       providerId: "console",
       upstreamId: "claude-sonnet-4.6",
       displayName: "Claude Sonnet 4.6",
-      aliases: [],
+      aliases: [
+        "claude-sonnet-4-6"
+      ],
       endpoints: [
         "chat"
       ],
@@ -28400,7 +29375,16 @@ var catalog_default = {
           "max"
         ],
         continuation: "opaque-provider-state",
-        defaultEffort: "high"
+        defaultEffort: "high",
+        effortRequest: {
+          value: {
+            field: "output_config.effort"
+          },
+          source: "official-doc",
+          confidence: "declared",
+          observedAt: "2026-09-25T12:30:00Z",
+          sourceRef: "https://platform.claude.com/docs/en/build-with-claude/effort — the effort page's supportedModels list names claude-fable-5-1, claude-fable-5, claude-opus-5-5, claude-opus-5, claude-opus-4-8, claude-opus-4-7, claude-opus-4-6, claude-opus-4-5-20251101, claude-sonnet-5 and claude-sonnet-4-6: effort is requested through `output_config.effort` (Claude 4.7 and later reject a manual `thinking.budget_tokens`, https://platform.claude.com/docs/en/build-with-claude/thinking-troubleshooting#rejected-configurations)."
+        }
       },
       pricing: {
         value: {
@@ -28416,6 +29400,13 @@ var catalog_default = {
       },
       unsupportedParameters: [],
       status: "candidate",
+      deferredToolLoading: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-25T18:30:00Z",
+        sourceRef: 'https://platform.claude.com/docs/en/agents-and-tools/tool-use/tool-search-tool — the Model compatibility table lists Fable 5.1, Mythos 5.1, Fable 5, Mythos 5, Opus 5.5, Opus 5, Opus 4.8, 4.7, 4.6, Sonnet 4.6, Opus 4.5, Sonnet 4.5 and Haiku 4.5 (not Sonnet 5); "Custom tool search implementation": a client tool returns a tool_result whose content carries {type:"tool_reference", tool_name} blocks, every referenced tool declared in tools with defer_loading: true, and the API expands them; "The prefix is untouched, so prompt caching is preserved" (https://platform.claude.com/docs/en/agents-and-tools/tool-use/tool-use-with-prompt-caching).'
+      },
       canonicalModelId: "claude-sonnet-4.6",
       modelFamily: "claude"
     },
@@ -28557,6 +29548,15 @@ var catalog_default = {
           sourceRef: 'continuity report §4.4, BOTH halves. Own-state acceptance: §4.4 requires the model\'s own assistant blocks — `thinking` with its signature, `redacted_thinking` — replayed unchanged and in order across a tool loop. Why the domain is NARROW: Anthropic documents model switching as a boundary at which those blocks are STRIPPED, because they are tied to the producing model; "same provider" is therefore not "same continuation domain", and this domain is the model alone.',
           confidence: "declared",
           observedAt: "2026-09-16T00:00:00Z"
+        },
+        effortRequest: {
+          value: {
+            field: "output_config.effort"
+          },
+          source: "official-doc",
+          confidence: "declared",
+          observedAt: "2026-09-25T12:30:00Z",
+          sourceRef: "https://platform.claude.com/docs/en/build-with-claude/effort — the effort page's supportedModels list names claude-fable-5-1, claude-fable-5, claude-opus-5-5, claude-opus-5, claude-opus-4-8, claude-opus-4-7, claude-opus-4-6, claude-opus-4-5-20251101, claude-sonnet-5 and claude-sonnet-4-6: effort is requested through `output_config.effort` (Claude 4.7 and later reject a manual `thinking.budget_tokens`, https://platform.claude.com/docs/en/build-with-claude/thinking-troubleshooting#rejected-configurations)."
         }
       },
       pricing: {
@@ -28574,9 +29574,17 @@ var catalog_default = {
       unsupportedParameters: [
         "temperature",
         "top_p",
-        "top_k"
+        "top_k",
+        "thinking.type.enabled"
       ],
       status: "candidate",
+      assistantPrefill: {
+        value: false,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: `https://platform.claude.com/docs/en/models/opus-5-5/migration-guide — "Don't end messages with a prefilled assistant turn: it is rejected"; "Prefilling assistant messages returns a 400 error on Claude Opus 4.6 and later Opus models, including Claude Opus 5.5"; Sonnet 5 and Opus 5: "Manual extended thinking, non-default sampling parameters, and assistant prefill return a 400 error on both models"; the Fable 5.1 guide: prefill "returns a 400 error, unchanged from Claude Fable 5". Live gate 2026-09-26 on claude-opus-5-5: "This model does not support assistant message prefill. The conversation must end with a user message."`
+      },
       canonicalModelId: "claude-sonnet-5",
       modelFamily: "claude"
     },
@@ -29850,12 +30858,11 @@ var catalog_default = {
         supported: {
           value: true,
           source: "official-doc",
-          sourceRef: "https://api-docs.deepseek.com/guides/thinking_mode/ — Chat reasoning_effort and Responses reasoning.effort allow none/low/high/max; default high; none disables",
+          sourceRef: "https://api-docs.deepseek.com/guides/thinking_mode/ — Chat Completions thinking toggle uses thinking.type enabled/disabled; reasoning_effort low/high/max; default high. Responses reasoning.effort additionally accepts none",
           confidence: "declared",
-          observedAt: "2026-09-25T08:45:47Z"
+          observedAt: "2026-09-25T12:31:56.381Z"
         },
         efforts: [
-          "none",
           "low",
           "high",
           "max"
@@ -29980,12 +30987,11 @@ var catalog_default = {
         supported: {
           value: true,
           source: "official-doc",
-          sourceRef: "https://api-docs.deepseek.com/guides/thinking_mode/ — Chat reasoning_effort and Responses reasoning.effort allow none/low/high/max; default high; none disables",
+          sourceRef: "https://api-docs.deepseek.com/guides/thinking_mode/ — Chat Completions thinking toggle uses thinking.type enabled/disabled; reasoning_effort low/high/max; default high. Responses reasoning.effort additionally accepts none",
           confidence: "declared",
-          observedAt: "2026-09-25T08:45:47Z"
+          observedAt: "2026-09-25T12:31:56.381Z"
         },
         efforts: [
-          "none",
           "low",
           "high",
           "max"
@@ -43177,9 +44183,9 @@ var catalog_default = {
           cacheWritePerMTokUsd: 0.375
         },
         source: "official-doc",
-        sourceRef: "https://platform.minimax.io/subscribe/token-plan?tab=api-enterprise — current MiniMax API pay-as-you-go Token Plan price table for this M2.7 model, per million tokens (retrieved 2026-09-19).",
+        sourceRef: "https://platform.minimax.io/docs/guides/pricing-paygo — MiniMax-M2.7 Pay as You Go Standard table: $0.3 input/$1.2 output/$0.06 prompt-cache read/$0.375 prompt-cache write per million tokens",
         confidence: "declared",
-        observedAt: "2026-09-19T00:00:00Z"
+        observedAt: "2026-09-25T12:31:56.381Z"
       },
       unsupportedParameters: [
         "function_call"
@@ -43269,9 +44275,9 @@ var catalog_default = {
           cacheWritePerMTokUsd: 0.375
         },
         source: "official-doc",
-        sourceRef: "https://platform.minimax.io/subscribe/token-plan?tab=api-enterprise — current MiniMax API pay-as-you-go Token Plan price table for this M2.7 model, per million tokens (retrieved 2026-09-19).",
+        sourceRef: "https://platform.minimax.io/docs/guides/pricing-paygo — MiniMax-M2.7-highspeed Pay as You Go Standard table: $0.6 input/$2.4 output/$0.06 prompt-cache read/$0.375 prompt-cache write per million tokens",
         confidence: "declared",
-        observedAt: "2026-09-19T00:00:00Z"
+        observedAt: "2026-09-25T12:31:56.381Z"
       },
       unsupportedParameters: [
         "function_call"
@@ -50879,6 +51885,13 @@ var catalog_default = {
       },
       unsupportedParameters: [],
       status: "candidate",
+      promptCacheKey: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-25T19:30:00Z",
+        sourceRef: 'https://developers.openai.com/api/docs/guides/prompt-caching — the Responses API takes prompt_cache_key: "Use a stable prompt_cache_key to optimize cache routing for requests that share a reusable prefix" (pre-GPT-5.6); on GPT-5.6+ routing is automatic and the key is "optional for separate cache accounting".'
+      },
       canonicalModelId: "gpt-4.1",
       modelFamily: "gpt"
     },
@@ -50959,6 +51972,13 @@ var catalog_default = {
       },
       unsupportedParameters: [],
       status: "candidate",
+      promptCacheKey: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-25T19:30:00Z",
+        sourceRef: 'https://developers.openai.com/api/docs/guides/prompt-caching — the Responses API takes prompt_cache_key: "Use a stable prompt_cache_key to optimize cache routing for requests that share a reusable prefix" (pre-GPT-5.6); on GPT-5.6+ routing is automatic and the key is "optional for separate cache accounting".'
+      },
       canonicalModelId: "gpt-4.1-mini",
       modelFamily: "gpt"
     },
@@ -51046,6 +52066,13 @@ var catalog_default = {
       },
       unsupportedParameters: [],
       status: "candidate",
+      promptCacheKey: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-25T19:30:00Z",
+        sourceRef: 'https://developers.openai.com/api/docs/guides/prompt-caching — the Responses API takes prompt_cache_key: "Use a stable prompt_cache_key to optimize cache routing for requests that share a reusable prefix" (pre-GPT-5.6); on GPT-5.6+ routing is automatic and the key is "optional for separate cache accounting".'
+      },
       canonicalModelId: "gpt-4.1-nano",
       modelFamily: "gpt"
     },
@@ -51126,6 +52153,13 @@ var catalog_default = {
       },
       unsupportedParameters: [],
       status: "candidate",
+      promptCacheKey: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-25T19:30:00Z",
+        sourceRef: 'https://developers.openai.com/api/docs/guides/prompt-caching — the Responses API takes prompt_cache_key: "Use a stable prompt_cache_key to optimize cache routing for requests that share a reusable prefix" (pre-GPT-5.6); on GPT-5.6+ routing is automatic and the key is "optional for separate cache accounting".'
+      },
       canonicalModelId: "gpt-4o",
       modelFamily: "gpt"
     },
@@ -51206,6 +52240,13 @@ var catalog_default = {
       },
       unsupportedParameters: [],
       status: "candidate",
+      promptCacheKey: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-25T19:30:00Z",
+        sourceRef: 'https://developers.openai.com/api/docs/guides/prompt-caching — the Responses API takes prompt_cache_key: "Use a stable prompt_cache_key to optimize cache routing for requests that share a reusable prefix" (pre-GPT-5.6); on GPT-5.6+ routing is automatic and the key is "optional for separate cache accounting".'
+      },
       canonicalModelId: "gpt-4o-2024-11-20",
       modelFamily: "gpt"
     },
@@ -51286,6 +52327,13 @@ var catalog_default = {
       },
       unsupportedParameters: [],
       status: "candidate",
+      promptCacheKey: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-25T19:30:00Z",
+        sourceRef: 'https://developers.openai.com/api/docs/guides/prompt-caching — the Responses API takes prompt_cache_key: "Use a stable prompt_cache_key to optimize cache routing for requests that share a reusable prefix" (pre-GPT-5.6); on GPT-5.6+ routing is automatic and the key is "optional for separate cache accounting".'
+      },
       canonicalModelId: "gpt-4o-mini",
       modelFamily: "gpt"
     },
@@ -51391,6 +52439,34 @@ var catalog_default = {
       },
       unsupportedParameters: [],
       status: "candidate",
+      promptCacheKey: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-25T19:30:00Z",
+        sourceRef: 'https://developers.openai.com/api/docs/guides/prompt-caching — the Responses API takes prompt_cache_key: "Use a stable prompt_cache_key to optimize cache routing for requests that share a reusable prefix" (pre-GPT-5.6); on GPT-5.6+ routing is automatic and the key is "optional for separate cache accounting".'
+      },
+      clientToolSearch: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: 'https://developers.openai.com/api/docs/guides/tools-tool-search — "Only gpt-5.4 and later models support tool_search" in the Responses API; client execution: {"type":"tool_search","execution":"client",description,parameters}, tool_search_call -> tool_search_output {call_id, execution:"client", status, tools}, deferred functions keep defer_loading:true, optionally inside {"type":"namespace",name,description,tools}. Retrieved 2026-09-26.'
+      },
+      additionalToolsItem: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: 'https://developers.openai.com/api/docs/guides/tools-tool-search — {"type":"additional_tools","role":"developer","tools":[...]}: "Tools in an additional_tools item become available only after that item appears in the input"; its position is kept on replay. The page names no model list, so it is set only on the rows the same page documents tool_search for (GPT-5.4 and later). NOT on codex-oauth: codex-rs sends one only at input[0] (core/src/client.rs:910-941), never mid-conversation -- the live probe decides. Retrieved 2026-09-26.'
+      },
+      allowedToolsChoice: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: 'https://developers.openai.com/api/docs/guides/function-calling — tool_choice {"type":"allowed_tools","mode":"auto"|"required","tools":[{"type":"function","name":…}]}: "make only a subset of tools available across model requests, but not modify the list of tools you pass in, so you can maximize savings from prompt caching"; "When you use tool search, tool_choice still applies to the tools that are currently callable in the turn." No model list on the page; set on the same GPT-5.4+ rows. NOT on codex-oauth: codex-rs always sends tool_choice "auto" (core/src/client.rs:1004) and its allowed_tools is a client-side filter (ext/extension-api/src/tool_policy.rs:14-36) -- the live probe decides. Retrieved 2026-09-26.'
+      },
       canonicalModelId: "gpt-5.4",
       modelFamily: "gpt"
     },
@@ -51503,6 +52579,34 @@ var catalog_default = {
       },
       unsupportedParameters: [],
       status: "candidate",
+      promptCacheKey: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-25T19:30:00Z",
+        sourceRef: 'https://developers.openai.com/api/docs/guides/prompt-caching — the Responses API takes prompt_cache_key: "Use a stable prompt_cache_key to optimize cache routing for requests that share a reusable prefix" (pre-GPT-5.6); on GPT-5.6+ routing is automatic and the key is "optional for separate cache accounting".'
+      },
+      clientToolSearch: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: 'https://developers.openai.com/api/docs/guides/tools-tool-search — "Only gpt-5.4 and later models support tool_search" in the Responses API; client execution: {"type":"tool_search","execution":"client",description,parameters}, tool_search_call -> tool_search_output {call_id, execution:"client", status, tools}, deferred functions keep defer_loading:true, optionally inside {"type":"namespace",name,description,tools}. Retrieved 2026-09-26.'
+      },
+      additionalToolsItem: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: 'https://developers.openai.com/api/docs/guides/tools-tool-search — {"type":"additional_tools","role":"developer","tools":[...]}: "Tools in an additional_tools item become available only after that item appears in the input"; its position is kept on replay. The page names no model list, so it is set only on the rows the same page documents tool_search for (GPT-5.4 and later). NOT on codex-oauth: codex-rs sends one only at input[0] (core/src/client.rs:910-941), never mid-conversation -- the live probe decides. Retrieved 2026-09-26.'
+      },
+      allowedToolsChoice: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: 'https://developers.openai.com/api/docs/guides/function-calling — tool_choice {"type":"allowed_tools","mode":"auto"|"required","tools":[{"type":"function","name":…}]}: "make only a subset of tools available across model requests, but not modify the list of tools you pass in, so you can maximize savings from prompt caching"; "When you use tool search, tool_choice still applies to the tools that are currently callable in the turn." No model list on the page; set on the same GPT-5.4+ rows. NOT on codex-oauth: codex-rs always sends tool_choice "auto" (core/src/client.rs:1004) and its allowed_tools is a client-side filter (ext/extension-api/src/tool_policy.rs:14-36) -- the live probe decides. Retrieved 2026-09-26.'
+      },
       canonicalModelId: "gpt-5.4-mini",
       modelFamily: "gpt"
     },
@@ -51615,6 +52719,34 @@ var catalog_default = {
       },
       unsupportedParameters: [],
       status: "candidate",
+      promptCacheKey: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-25T19:30:00Z",
+        sourceRef: 'https://developers.openai.com/api/docs/guides/prompt-caching — the Responses API takes prompt_cache_key: "Use a stable prompt_cache_key to optimize cache routing for requests that share a reusable prefix" (pre-GPT-5.6); on GPT-5.6+ routing is automatic and the key is "optional for separate cache accounting".'
+      },
+      clientToolSearch: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: 'https://developers.openai.com/api/docs/guides/tools-tool-search — "Only gpt-5.4 and later models support tool_search" in the Responses API; client execution: {"type":"tool_search","execution":"client",description,parameters}, tool_search_call -> tool_search_output {call_id, execution:"client", status, tools}, deferred functions keep defer_loading:true, optionally inside {"type":"namespace",name,description,tools}. Retrieved 2026-09-26.'
+      },
+      additionalToolsItem: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: 'https://developers.openai.com/api/docs/guides/tools-tool-search — {"type":"additional_tools","role":"developer","tools":[...]}: "Tools in an additional_tools item become available only after that item appears in the input"; its position is kept on replay. The page names no model list, so it is set only on the rows the same page documents tool_search for (GPT-5.4 and later). NOT on codex-oauth: codex-rs sends one only at input[0] (core/src/client.rs:910-941), never mid-conversation -- the live probe decides. Retrieved 2026-09-26.'
+      },
+      allowedToolsChoice: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: 'https://developers.openai.com/api/docs/guides/function-calling — tool_choice {"type":"allowed_tools","mode":"auto"|"required","tools":[{"type":"function","name":…}]}: "make only a subset of tools available across model requests, but not modify the list of tools you pass in, so you can maximize savings from prompt caching"; "When you use tool search, tool_choice still applies to the tools that are currently callable in the turn." No model list on the page; set on the same GPT-5.4+ rows. NOT on codex-oauth: codex-rs always sends tool_choice "auto" (core/src/client.rs:1004) and its allowed_tools is a client-side filter (ext/extension-api/src/tool_policy.rs:14-36) -- the live probe decides. Retrieved 2026-09-26.'
+      },
       canonicalModelId: "gpt-5.4-nano",
       modelFamily: "gpt"
     },
@@ -51702,6 +52834,34 @@ var catalog_default = {
       },
       unsupportedParameters: [],
       status: "candidate",
+      promptCacheKey: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-25T19:30:00Z",
+        sourceRef: 'https://developers.openai.com/api/docs/guides/prompt-caching — the Responses API takes prompt_cache_key: "Use a stable prompt_cache_key to optimize cache routing for requests that share a reusable prefix" (pre-GPT-5.6); on GPT-5.6+ routing is automatic and the key is "optional for separate cache accounting".'
+      },
+      clientToolSearch: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: 'https://developers.openai.com/api/docs/guides/tools-tool-search — "Only gpt-5.4 and later models support tool_search" in the Responses API; client execution: {"type":"tool_search","execution":"client",description,parameters}, tool_search_call -> tool_search_output {call_id, execution:"client", status, tools}, deferred functions keep defer_loading:true, optionally inside {"type":"namespace",name,description,tools}. Retrieved 2026-09-26.'
+      },
+      additionalToolsItem: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: 'https://developers.openai.com/api/docs/guides/tools-tool-search — {"type":"additional_tools","role":"developer","tools":[...]}: "Tools in an additional_tools item become available only after that item appears in the input"; its position is kept on replay. The page names no model list, so it is set only on the rows the same page documents tool_search for (GPT-5.4 and later). NOT on codex-oauth: codex-rs sends one only at input[0] (core/src/client.rs:910-941), never mid-conversation -- the live probe decides. Retrieved 2026-09-26.'
+      },
+      allowedToolsChoice: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: 'https://developers.openai.com/api/docs/guides/function-calling — tool_choice {"type":"allowed_tools","mode":"auto"|"required","tools":[{"type":"function","name":…}]}: "make only a subset of tools available across model requests, but not modify the list of tools you pass in, so you can maximize savings from prompt caching"; "When you use tool search, tool_choice still applies to the tools that are currently callable in the turn." No model list on the page; set on the same GPT-5.4+ rows. NOT on codex-oauth: codex-rs always sends tool_choice "auto" (core/src/client.rs:1004) and its allowed_tools is a client-side filter (ext/extension-api/src/tool_policy.rs:14-36) -- the live probe decides. Retrieved 2026-09-26.'
+      },
       canonicalModelId: "gpt-5.4-pro",
       modelFamily: "gpt"
     },
@@ -51807,6 +52967,34 @@ var catalog_default = {
       },
       unsupportedParameters: [],
       status: "candidate",
+      promptCacheKey: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-25T19:30:00Z",
+        sourceRef: 'https://developers.openai.com/api/docs/guides/prompt-caching — the Responses API takes prompt_cache_key: "Use a stable prompt_cache_key to optimize cache routing for requests that share a reusable prefix" (pre-GPT-5.6); on GPT-5.6+ routing is automatic and the key is "optional for separate cache accounting".'
+      },
+      clientToolSearch: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: 'https://developers.openai.com/api/docs/guides/tools-tool-search — "Only gpt-5.4 and later models support tool_search" in the Responses API; client execution: {"type":"tool_search","execution":"client",description,parameters}, tool_search_call -> tool_search_output {call_id, execution:"client", status, tools}, deferred functions keep defer_loading:true, optionally inside {"type":"namespace",name,description,tools}. Retrieved 2026-09-26.'
+      },
+      additionalToolsItem: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: 'https://developers.openai.com/api/docs/guides/tools-tool-search — {"type":"additional_tools","role":"developer","tools":[...]}: "Tools in an additional_tools item become available only after that item appears in the input"; its position is kept on replay. The page names no model list, so it is set only on the rows the same page documents tool_search for (GPT-5.4 and later). NOT on codex-oauth: codex-rs sends one only at input[0] (core/src/client.rs:910-941), never mid-conversation -- the live probe decides. Retrieved 2026-09-26.'
+      },
+      allowedToolsChoice: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: 'https://developers.openai.com/api/docs/guides/function-calling — tool_choice {"type":"allowed_tools","mode":"auto"|"required","tools":[{"type":"function","name":…}]}: "make only a subset of tools available across model requests, but not modify the list of tools you pass in, so you can maximize savings from prompt caching"; "When you use tool search, tool_choice still applies to the tools that are currently callable in the turn." No model list on the page; set on the same GPT-5.4+ rows. NOT on codex-oauth: codex-rs always sends tool_choice "auto" (core/src/client.rs:1004) and its allowed_tools is a client-side filter (ext/extension-api/src/tool_policy.rs:14-36) -- the live probe decides. Retrieved 2026-09-26.'
+      },
       canonicalModelId: "gpt-5.5",
       modelFamily: "gpt"
     },
@@ -51901,6 +53089,34 @@ var catalog_default = {
       },
       unsupportedParameters: [],
       status: "candidate",
+      promptCacheKey: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-25T19:30:00Z",
+        sourceRef: 'https://developers.openai.com/api/docs/guides/prompt-caching — the Responses API takes prompt_cache_key: "Use a stable prompt_cache_key to optimize cache routing for requests that share a reusable prefix" (pre-GPT-5.6); on GPT-5.6+ routing is automatic and the key is "optional for separate cache accounting".'
+      },
+      clientToolSearch: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: 'https://developers.openai.com/api/docs/guides/tools-tool-search — "Only gpt-5.4 and later models support tool_search" in the Responses API; client execution: {"type":"tool_search","execution":"client",description,parameters}, tool_search_call -> tool_search_output {call_id, execution:"client", status, tools}, deferred functions keep defer_loading:true, optionally inside {"type":"namespace",name,description,tools}. Retrieved 2026-09-26.'
+      },
+      additionalToolsItem: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: 'https://developers.openai.com/api/docs/guides/tools-tool-search — {"type":"additional_tools","role":"developer","tools":[...]}: "Tools in an additional_tools item become available only after that item appears in the input"; its position is kept on replay. The page names no model list, so it is set only on the rows the same page documents tool_search for (GPT-5.4 and later). NOT on codex-oauth: codex-rs sends one only at input[0] (core/src/client.rs:910-941), never mid-conversation -- the live probe decides. Retrieved 2026-09-26.'
+      },
+      allowedToolsChoice: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: 'https://developers.openai.com/api/docs/guides/function-calling — tool_choice {"type":"allowed_tools","mode":"auto"|"required","tools":[{"type":"function","name":…}]}: "make only a subset of tools available across model requests, but not modify the list of tools you pass in, so you can maximize savings from prompt caching"; "When you use tool search, tool_choice still applies to the tools that are currently callable in the turn." No model list on the page; set on the same GPT-5.4+ rows. NOT on codex-oauth: codex-rs always sends tool_choice "auto" (core/src/client.rs:1004) and its allowed_tools is a client-side filter (ext/extension-api/src/tool_policy.rs:14-36) -- the live probe decides. Retrieved 2026-09-26.'
+      },
       canonicalModelId: "gpt-5.5-pro",
       modelFamily: "gpt"
     },
@@ -52015,6 +53231,34 @@ var catalog_default = {
       },
       unsupportedParameters: [],
       status: "candidate",
+      promptCacheKey: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-25T19:30:00Z",
+        sourceRef: 'https://developers.openai.com/api/docs/guides/prompt-caching — the Responses API takes prompt_cache_key: "Use a stable prompt_cache_key to optimize cache routing for requests that share a reusable prefix" (pre-GPT-5.6); on GPT-5.6+ routing is automatic and the key is "optional for separate cache accounting".'
+      },
+      clientToolSearch: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: 'https://developers.openai.com/api/docs/guides/tools-tool-search — "Only gpt-5.4 and later models support tool_search" in the Responses API; client execution: {"type":"tool_search","execution":"client",description,parameters}, tool_search_call -> tool_search_output {call_id, execution:"client", status, tools}, deferred functions keep defer_loading:true, optionally inside {"type":"namespace",name,description,tools}. Retrieved 2026-09-26.'
+      },
+      additionalToolsItem: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: 'https://developers.openai.com/api/docs/guides/tools-tool-search — {"type":"additional_tools","role":"developer","tools":[...]}: "Tools in an additional_tools item become available only after that item appears in the input"; its position is kept on replay. The page names no model list, so it is set only on the rows the same page documents tool_search for (GPT-5.4 and later). NOT on codex-oauth: codex-rs sends one only at input[0] (core/src/client.rs:910-941), never mid-conversation -- the live probe decides. Retrieved 2026-09-26.'
+      },
+      allowedToolsChoice: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: 'https://developers.openai.com/api/docs/guides/function-calling — tool_choice {"type":"allowed_tools","mode":"auto"|"required","tools":[{"type":"function","name":…}]}: "make only a subset of tools available across model requests, but not modify the list of tools you pass in, so you can maximize savings from prompt caching"; "When you use tool search, tool_choice still applies to the tools that are currently callable in the turn." No model list on the page; set on the same GPT-5.4+ rows. NOT on codex-oauth: codex-rs always sends tool_choice "auto" (core/src/client.rs:1004) and its allowed_tools is a client-side filter (ext/extension-api/src/tool_policy.rs:14-36) -- the live probe decides. Retrieved 2026-09-26.'
+      },
       canonicalModelId: "gpt-5.6",
       modelFamily: "gpt"
     },
@@ -52129,6 +53373,34 @@ var catalog_default = {
       },
       unsupportedParameters: [],
       status: "candidate",
+      promptCacheKey: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-25T19:30:00Z",
+        sourceRef: 'https://developers.openai.com/api/docs/guides/prompt-caching — the Responses API takes prompt_cache_key: "Use a stable prompt_cache_key to optimize cache routing for requests that share a reusable prefix" (pre-GPT-5.6); on GPT-5.6+ routing is automatic and the key is "optional for separate cache accounting".'
+      },
+      clientToolSearch: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: 'https://developers.openai.com/api/docs/guides/tools-tool-search — "Only gpt-5.4 and later models support tool_search" in the Responses API; client execution: {"type":"tool_search","execution":"client",description,parameters}, tool_search_call -> tool_search_output {call_id, execution:"client", status, tools}, deferred functions keep defer_loading:true, optionally inside {"type":"namespace",name,description,tools}. Retrieved 2026-09-26.'
+      },
+      additionalToolsItem: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: 'https://developers.openai.com/api/docs/guides/tools-tool-search — {"type":"additional_tools","role":"developer","tools":[...]}: "Tools in an additional_tools item become available only after that item appears in the input"; its position is kept on replay. The page names no model list, so it is set only on the rows the same page documents tool_search for (GPT-5.4 and later). NOT on codex-oauth: codex-rs sends one only at input[0] (core/src/client.rs:910-941), never mid-conversation -- the live probe decides. Retrieved 2026-09-26.'
+      },
+      allowedToolsChoice: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: 'https://developers.openai.com/api/docs/guides/function-calling — tool_choice {"type":"allowed_tools","mode":"auto"|"required","tools":[{"type":"function","name":…}]}: "make only a subset of tools available across model requests, but not modify the list of tools you pass in, so you can maximize savings from prompt caching"; "When you use tool search, tool_choice still applies to the tools that are currently callable in the turn." No model list on the page; set on the same GPT-5.4+ rows. NOT on codex-oauth: codex-rs always sends tool_choice "auto" (core/src/client.rs:1004) and its allowed_tools is a client-side filter (ext/extension-api/src/tool_policy.rs:14-36) -- the live probe decides. Retrieved 2026-09-26.'
+      },
       canonicalModelId: "gpt-5.6-luna",
       modelFamily: "gpt"
     },
@@ -52243,6 +53515,34 @@ var catalog_default = {
       },
       unsupportedParameters: [],
       status: "candidate",
+      promptCacheKey: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-25T19:30:00Z",
+        sourceRef: 'https://developers.openai.com/api/docs/guides/prompt-caching — the Responses API takes prompt_cache_key: "Use a stable prompt_cache_key to optimize cache routing for requests that share a reusable prefix" (pre-GPT-5.6); on GPT-5.6+ routing is automatic and the key is "optional for separate cache accounting".'
+      },
+      clientToolSearch: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: 'https://developers.openai.com/api/docs/guides/tools-tool-search — "Only gpt-5.4 and later models support tool_search" in the Responses API; client execution: {"type":"tool_search","execution":"client",description,parameters}, tool_search_call -> tool_search_output {call_id, execution:"client", status, tools}, deferred functions keep defer_loading:true, optionally inside {"type":"namespace",name,description,tools}. Retrieved 2026-09-26.'
+      },
+      additionalToolsItem: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: 'https://developers.openai.com/api/docs/guides/tools-tool-search — {"type":"additional_tools","role":"developer","tools":[...]}: "Tools in an additional_tools item become available only after that item appears in the input"; its position is kept on replay. The page names no model list, so it is set only on the rows the same page documents tool_search for (GPT-5.4 and later). NOT on codex-oauth: codex-rs sends one only at input[0] (core/src/client.rs:910-941), never mid-conversation -- the live probe decides. Retrieved 2026-09-26.'
+      },
+      allowedToolsChoice: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: 'https://developers.openai.com/api/docs/guides/function-calling — tool_choice {"type":"allowed_tools","mode":"auto"|"required","tools":[{"type":"function","name":…}]}: "make only a subset of tools available across model requests, but not modify the list of tools you pass in, so you can maximize savings from prompt caching"; "When you use tool search, tool_choice still applies to the tools that are currently callable in the turn." No model list on the page; set on the same GPT-5.4+ rows. NOT on codex-oauth: codex-rs always sends tool_choice "auto" (core/src/client.rs:1004) and its allowed_tools is a client-side filter (ext/extension-api/src/tool_policy.rs:14-36) -- the live probe decides. Retrieved 2026-09-26.'
+      },
       canonicalModelId: "gpt-5.6-sol",
       modelFamily: "gpt"
     },
@@ -52357,6 +53657,34 @@ var catalog_default = {
       },
       unsupportedParameters: [],
       status: "candidate",
+      promptCacheKey: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-25T19:30:00Z",
+        sourceRef: 'https://developers.openai.com/api/docs/guides/prompt-caching — the Responses API takes prompt_cache_key: "Use a stable prompt_cache_key to optimize cache routing for requests that share a reusable prefix" (pre-GPT-5.6); on GPT-5.6+ routing is automatic and the key is "optional for separate cache accounting".'
+      },
+      clientToolSearch: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: 'https://developers.openai.com/api/docs/guides/tools-tool-search — "Only gpt-5.4 and later models support tool_search" in the Responses API; client execution: {"type":"tool_search","execution":"client",description,parameters}, tool_search_call -> tool_search_output {call_id, execution:"client", status, tools}, deferred functions keep defer_loading:true, optionally inside {"type":"namespace",name,description,tools}. Retrieved 2026-09-26.'
+      },
+      additionalToolsItem: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: 'https://developers.openai.com/api/docs/guides/tools-tool-search — {"type":"additional_tools","role":"developer","tools":[...]}: "Tools in an additional_tools item become available only after that item appears in the input"; its position is kept on replay. The page names no model list, so it is set only on the rows the same page documents tool_search for (GPT-5.4 and later). NOT on codex-oauth: codex-rs sends one only at input[0] (core/src/client.rs:910-941), never mid-conversation -- the live probe decides. Retrieved 2026-09-26.'
+      },
+      allowedToolsChoice: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: 'https://developers.openai.com/api/docs/guides/function-calling — tool_choice {"type":"allowed_tools","mode":"auto"|"required","tools":[{"type":"function","name":…}]}: "make only a subset of tools available across model requests, but not modify the list of tools you pass in, so you can maximize savings from prompt caching"; "When you use tool search, tool_choice still applies to the tools that are currently callable in the turn." No model list on the page; set on the same GPT-5.4+ rows. NOT on codex-oauth: codex-rs always sends tool_choice "auto" (core/src/client.rs:1004) and its allowed_tools is a client-side filter (ext/extension-api/src/tool_policy.rs:14-36) -- the live probe decides. Retrieved 2026-09-26.'
+      },
       canonicalModelId: "gpt-5.6-terra",
       modelFamily: "gpt"
     },
@@ -52454,7 +53782,7 @@ var catalog_default = {
           "max"
         ],
         continuation: "opaque-provider-state",
-        defaultEffort: "low",
+        defaultEffort: "medium",
         readableState: {
           value: "summary",
           source: "official-doc",
@@ -52491,6 +53819,15 @@ var catalog_default = {
           sourceRef: 'continuity report §5.1/§5.3 for own-state acceptance — the Responses surface replays its own completed `reasoning.encrypted_content` items to the same model.  WS-13 §8.2: "OpenAI-compatible" is never the capability test, so this key shares NO domain with any other row.',
           confidence: "declared",
           observedAt: "2026-09-06T00:00:00Z"
+        },
+        perMessageEffort: {
+          value: {
+            item: "configuration_update"
+          },
+          source: "official-doc",
+          confidence: "declared",
+          observedAt: "2026-09-26T00:00:00Z",
+          sourceRef: `https://developers.openai.com/api/docs/guides/reasoning — "Configuration updates are supported by the GPT-6 model family in standard, single-agent mode. They change only reasoning effort." Shape {"type":"configuration_update","reasoning":{"effort":…}}, placed "before the next user message in the input array"; "the API rejects adjacent updates"; "The response's reasoning.effort continues to report the request-level setting"; with store:false, replay updates "in their original positions". Retrieved 2026-09-26. NOT set on codex-oauth/gpt-6-* (codex-rs enables it only behind an off-by-default feature, backend acceptance unprobed — scripts/probe-openai-midconv.ts decides) nor on azure-openai (undocumented there).`
         }
       },
       pricing: {
@@ -52507,6 +53844,34 @@ var catalog_default = {
       },
       unsupportedParameters: [],
       status: "candidate",
+      promptCacheKey: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-25T19:30:00Z",
+        sourceRef: 'https://developers.openai.com/api/docs/guides/prompt-caching — the Responses API takes prompt_cache_key: "Use a stable prompt_cache_key to optimize cache routing for requests that share a reusable prefix" (pre-GPT-5.6); on GPT-5.6+ routing is automatic and the key is "optional for separate cache accounting".'
+      },
+      clientToolSearch: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: 'https://developers.openai.com/api/docs/guides/tools-tool-search — "Only gpt-5.4 and later models support tool_search" in the Responses API; client execution: {"type":"tool_search","execution":"client",description,parameters}, tool_search_call -> tool_search_output {call_id, execution:"client", status, tools}, deferred functions keep defer_loading:true, optionally inside {"type":"namespace",name,description,tools}. Retrieved 2026-09-26.'
+      },
+      additionalToolsItem: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: 'https://developers.openai.com/api/docs/guides/tools-tool-search — {"type":"additional_tools","role":"developer","tools":[...]}: "Tools in an additional_tools item become available only after that item appears in the input"; its position is kept on replay. The page names no model list, so it is set only on the rows the same page documents tool_search for (GPT-5.4 and later). NOT on codex-oauth: codex-rs sends one only at input[0] (core/src/client.rs:910-941), never mid-conversation -- the live probe decides. Retrieved 2026-09-26.'
+      },
+      allowedToolsChoice: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: 'https://developers.openai.com/api/docs/guides/function-calling — tool_choice {"type":"allowed_tools","mode":"auto"|"required","tools":[{"type":"function","name":…}]}: "make only a subset of tools available across model requests, but not modify the list of tools you pass in, so you can maximize savings from prompt caching"; "When you use tool search, tool_choice still applies to the tools that are currently callable in the turn." No model list on the page; set on the same GPT-5.4+ rows. NOT on codex-oauth: codex-rs always sends tool_choice "auto" (core/src/client.rs:1004) and its allowed_tools is a client-side filter (ext/extension-api/src/tool_policy.rs:14-36) -- the live probe decides. Retrieved 2026-09-26.'
+      },
       canonicalModelId: "gpt-6-astra",
       modelFamily: "gpt"
     },
@@ -52642,6 +54007,15 @@ var catalog_default = {
           sourceRef: 'continuity report §5.1/§5.3 for own-state acceptance — the Responses surface replays its own completed `reasoning.encrypted_content` items to the same model.  WS-13 §8.2: "OpenAI-compatible" is never the capability test, so this key shares NO domain with any other row.',
           confidence: "declared",
           observedAt: "2026-09-06T00:00:00Z"
+        },
+        perMessageEffort: {
+          value: {
+            item: "configuration_update"
+          },
+          source: "official-doc",
+          confidence: "declared",
+          observedAt: "2026-09-26T00:00:00Z",
+          sourceRef: `https://developers.openai.com/api/docs/guides/reasoning — "Configuration updates are supported by the GPT-6 model family in standard, single-agent mode. They change only reasoning effort." Shape {"type":"configuration_update","reasoning":{"effort":…}}, placed "before the next user message in the input array"; "the API rejects adjacent updates"; "The response's reasoning.effort continues to report the request-level setting"; with store:false, replay updates "in their original positions". Retrieved 2026-09-26. NOT set on codex-oauth/gpt-6-* (codex-rs enables it only behind an off-by-default feature, backend acceptance unprobed — scripts/probe-openai-midconv.ts decides) nor on azure-openai (undocumented there).`
         }
       },
       pricing: {
@@ -52658,6 +54032,34 @@ var catalog_default = {
       },
       unsupportedParameters: [],
       status: "candidate",
+      promptCacheKey: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-25T19:30:00Z",
+        sourceRef: 'https://developers.openai.com/api/docs/guides/prompt-caching — the Responses API takes prompt_cache_key: "Use a stable prompt_cache_key to optimize cache routing for requests that share a reusable prefix" (pre-GPT-5.6); on GPT-5.6+ routing is automatic and the key is "optional for separate cache accounting".'
+      },
+      clientToolSearch: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: 'https://developers.openai.com/api/docs/guides/tools-tool-search — "Only gpt-5.4 and later models support tool_search" in the Responses API; client execution: {"type":"tool_search","execution":"client",description,parameters}, tool_search_call -> tool_search_output {call_id, execution:"client", status, tools}, deferred functions keep defer_loading:true, optionally inside {"type":"namespace",name,description,tools}. Retrieved 2026-09-26.'
+      },
+      additionalToolsItem: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: 'https://developers.openai.com/api/docs/guides/tools-tool-search — {"type":"additional_tools","role":"developer","tools":[...]}: "Tools in an additional_tools item become available only after that item appears in the input"; its position is kept on replay. The page names no model list, so it is set only on the rows the same page documents tool_search for (GPT-5.4 and later). NOT on codex-oauth: codex-rs sends one only at input[0] (core/src/client.rs:910-941), never mid-conversation -- the live probe decides. Retrieved 2026-09-26.'
+      },
+      allowedToolsChoice: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: 'https://developers.openai.com/api/docs/guides/function-calling — tool_choice {"type":"allowed_tools","mode":"auto"|"required","tools":[{"type":"function","name":…}]}: "make only a subset of tools available across model requests, but not modify the list of tools you pass in, so you can maximize savings from prompt caching"; "When you use tool search, tool_choice still applies to the tools that are currently callable in the turn." No model list on the page; set on the same GPT-5.4+ rows. NOT on codex-oauth: codex-rs always sends tool_choice "auto" (core/src/client.rs:1004) and its allowed_tools is a client-side filter (ext/extension-api/src/tool_policy.rs:14-36) -- the live probe decides. Retrieved 2026-09-26.'
+      },
       canonicalModelId: "gpt-6-luna",
       modelFamily: "gpt"
     },
@@ -52793,6 +54195,15 @@ var catalog_default = {
           sourceRef: 'continuity report §5.1/§5.3 for own-state acceptance — the Responses surface replays its own completed `reasoning.encrypted_content` items to the same model.  WS-13 §8.2: "OpenAI-compatible" is never the capability test, so this key shares NO domain with any other row.',
           confidence: "declared",
           observedAt: "2026-09-06T00:00:00Z"
+        },
+        perMessageEffort: {
+          value: {
+            item: "configuration_update"
+          },
+          source: "official-doc",
+          confidence: "declared",
+          observedAt: "2026-09-26T00:00:00Z",
+          sourceRef: `https://developers.openai.com/api/docs/guides/reasoning — "Configuration updates are supported by the GPT-6 model family in standard, single-agent mode. They change only reasoning effort." Shape {"type":"configuration_update","reasoning":{"effort":…}}, placed "before the next user message in the input array"; "the API rejects adjacent updates"; "The response's reasoning.effort continues to report the request-level setting"; with store:false, replay updates "in their original positions". Retrieved 2026-09-26. NOT set on codex-oauth/gpt-6-* (codex-rs enables it only behind an off-by-default feature, backend acceptance unprobed — scripts/probe-openai-midconv.ts decides) nor on azure-openai (undocumented there).`
         }
       },
       pricing: {
@@ -52809,6 +54220,34 @@ var catalog_default = {
       },
       unsupportedParameters: [],
       status: "candidate",
+      promptCacheKey: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-25T19:30:00Z",
+        sourceRef: 'https://developers.openai.com/api/docs/guides/prompt-caching — the Responses API takes prompt_cache_key: "Use a stable prompt_cache_key to optimize cache routing for requests that share a reusable prefix" (pre-GPT-5.6); on GPT-5.6+ routing is automatic and the key is "optional for separate cache accounting".'
+      },
+      clientToolSearch: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: 'https://developers.openai.com/api/docs/guides/tools-tool-search — "Only gpt-5.4 and later models support tool_search" in the Responses API; client execution: {"type":"tool_search","execution":"client",description,parameters}, tool_search_call -> tool_search_output {call_id, execution:"client", status, tools}, deferred functions keep defer_loading:true, optionally inside {"type":"namespace",name,description,tools}. Retrieved 2026-09-26.'
+      },
+      additionalToolsItem: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: 'https://developers.openai.com/api/docs/guides/tools-tool-search — {"type":"additional_tools","role":"developer","tools":[...]}: "Tools in an additional_tools item become available only after that item appears in the input"; its position is kept on replay. The page names no model list, so it is set only on the rows the same page documents tool_search for (GPT-5.4 and later). NOT on codex-oauth: codex-rs sends one only at input[0] (core/src/client.rs:910-941), never mid-conversation -- the live probe decides. Retrieved 2026-09-26.'
+      },
+      allowedToolsChoice: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-26T00:00:00Z",
+        sourceRef: 'https://developers.openai.com/api/docs/guides/function-calling — tool_choice {"type":"allowed_tools","mode":"auto"|"required","tools":[{"type":"function","name":…}]}: "make only a subset of tools available across model requests, but not modify the list of tools you pass in, so you can maximize savings from prompt caching"; "When you use tool search, tool_choice still applies to the tools that are currently callable in the turn." No model list on the page; set on the same GPT-5.4+ rows. NOT on codex-oauth: codex-rs always sends tool_choice "auto" (core/src/client.rs:1004) and its allowed_tools is a client-side filter (ext/extension-api/src/tool_policy.rs:14-36) -- the live probe decides. Retrieved 2026-09-26.'
+      },
       canonicalModelId: "gpt-6-sol",
       modelFamily: "gpt"
     },
@@ -52911,6 +54350,13 @@ var catalog_default = {
       },
       unsupportedParameters: [],
       status: "candidate",
+      promptCacheKey: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-25T19:30:00Z",
+        sourceRef: 'https://developers.openai.com/api/docs/guides/prompt-caching — the Responses API takes prompt_cache_key: "Use a stable prompt_cache_key to optimize cache routing for requests that share a reusable prefix" (pre-GPT-5.6); on GPT-5.6+ routing is automatic and the key is "optional for separate cache accounting".'
+      },
       canonicalModelId: "o3",
       modelFamily: "o-series"
     },
@@ -53005,6 +54451,13 @@ var catalog_default = {
       },
       unsupportedParameters: [],
       status: "candidate",
+      promptCacheKey: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-25T19:30:00Z",
+        sourceRef: 'https://developers.openai.com/api/docs/guides/prompt-caching — the Responses API takes prompt_cache_key: "Use a stable prompt_cache_key to optimize cache routing for requests that share a reusable prefix" (pre-GPT-5.6); on GPT-5.6+ routing is automatic and the key is "optional for separate cache accounting".'
+      },
       canonicalModelId: "o3-mini",
       modelFamily: "o-series"
     },
@@ -53093,6 +54546,7 @@ var catalog_default = {
           "high"
         ],
         continuation: "opaque-provider-state",
+        defaultEffort: "medium",
         readableState: {
           value: "summary",
           source: "official-doc",
@@ -53151,6 +54605,13 @@ var catalog_default = {
       },
       unsupportedParameters: [],
       status: "candidate",
+      promptCacheKey: {
+        value: true,
+        source: "official-doc",
+        confidence: "declared",
+        observedAt: "2026-09-25T19:30:00Z",
+        sourceRef: 'https://developers.openai.com/api/docs/guides/prompt-caching — the Responses API takes prompt_cache_key: "Use a stable prompt_cache_key to optimize cache routing for requests that share a reusable prefix" (pre-GPT-5.6); on GPT-5.6+ routing is automatic and the key is "optional for separate cache accounting".'
+      },
       canonicalModelId: "o4-mini",
       modelFamily: "o-series"
     },
@@ -66307,6 +67768,27 @@ var catalog_default = {
       endpoints: [
         "chat"
       ],
+      contextWindow: {
+        value: 1e6,
+        source: "official-doc",
+        sourceRef: "https://cloud.tencent.com/document/product/1823/130051 — deepseek-v4-flash-202605 row lists context window 1M tokens; k/M expanded as decimal",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
+      maxInputTokens: {
+        value: 1e6,
+        source: "official-doc",
+        sourceRef: "https://cloud.tencent.com/document/product/1823/130051 — deepseek-v4-flash-202605 row lists maximum input 1M tokens; k/M expanded as decimal",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
+      maxOutputTokens: {
+        value: 384000,
+        source: "official-doc",
+        sourceRef: "https://cloud.tencent.com/document/product/1823/130051 — deepseek-v4-flash-202605 row lists maximum output 384k tokens; k/M expanded as decimal",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
       inputModalities: {
         value: [
           "text"
@@ -66367,6 +67849,27 @@ var catalog_default = {
       endpoints: [
         "chat"
       ],
+      contextWindow: {
+        value: 1e6,
+        source: "official-doc",
+        sourceRef: "https://cloud.tencent.com/document/product/1823/130051 — deepseek-v4-pro-202606 row lists context window 1M tokens; k/M expanded as decimal",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
+      maxInputTokens: {
+        value: 1e6,
+        source: "official-doc",
+        sourceRef: "https://cloud.tencent.com/document/product/1823/130051 — deepseek-v4-pro-202606 row lists maximum input 1M tokens; k/M expanded as decimal",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
+      maxOutputTokens: {
+        value: 384000,
+        source: "official-doc",
+        sourceRef: "https://cloud.tencent.com/document/product/1823/130051 — deepseek-v4-pro-202606 row lists maximum output 384k tokens; k/M expanded as decimal",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
       inputModalities: {
         value: [
           "text"
@@ -66426,6 +67929,27 @@ var catalog_default = {
       endpoints: [
         "chat"
       ],
+      contextWindow: {
+        value: 200000,
+        source: "official-doc",
+        sourceRef: "https://cloud.tencent.com/document/product/1823/130051 — glm-5 row lists context window 200k tokens; k/M expanded as decimal",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
+      maxInputTokens: {
+        value: 200000,
+        source: "official-doc",
+        sourceRef: "https://cloud.tencent.com/document/product/1823/130051 — glm-5 row lists maximum input 200k tokens; k/M expanded as decimal",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
+      maxOutputTokens: {
+        value: 128000,
+        source: "official-doc",
+        sourceRef: "https://cloud.tencent.com/document/product/1823/130051 — glm-5 row lists maximum output 128k tokens; k/M expanded as decimal",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
       inputModalities: {
         value: [
           "text"
@@ -66485,6 +68009,27 @@ var catalog_default = {
       endpoints: [
         "chat"
       ],
+      contextWindow: {
+        value: 200000,
+        source: "official-doc",
+        sourceRef: "https://cloud.tencent.com/document/product/1823/130051 — glm-5.1 row lists context window 200k tokens; k/M expanded as decimal",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
+      maxInputTokens: {
+        value: 200000,
+        source: "official-doc",
+        sourceRef: "https://cloud.tencent.com/document/product/1823/130051 — glm-5.1 row lists maximum input 200k tokens; k/M expanded as decimal",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
+      maxOutputTokens: {
+        value: 128000,
+        source: "official-doc",
+        sourceRef: "https://cloud.tencent.com/document/product/1823/130051 — glm-5.1 row lists maximum output 128k tokens; k/M expanded as decimal",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
       inputModalities: {
         value: [
           "text"
@@ -66544,6 +68089,27 @@ var catalog_default = {
       endpoints: [
         "chat"
       ],
+      contextWindow: {
+        value: 1e6,
+        source: "official-doc",
+        sourceRef: "https://cloud.tencent.com/document/product/1823/130051 — glm-5.2 row lists context window 1M tokens; k/M expanded as decimal",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
+      maxInputTokens: {
+        value: 1e6,
+        source: "official-doc",
+        sourceRef: "https://cloud.tencent.com/document/product/1823/130051 — glm-5.2 row lists maximum input 1M tokens; k/M expanded as decimal",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
+      maxOutputTokens: {
+        value: 128000,
+        source: "official-doc",
+        sourceRef: "https://cloud.tencent.com/document/product/1823/130051 — glm-5.2 row lists maximum output 128k tokens; k/M expanded as decimal",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
       inputModalities: {
         value: [
           "text"
@@ -66603,6 +68169,27 @@ var catalog_default = {
       endpoints: [
         "chat"
       ],
+      contextWindow: {
+        value: 1e6,
+        source: "official-doc",
+        sourceRef: "https://cloud.tencent.com/document/product/1823/130051 — glm-5.3 row lists context window 1M tokens; k/M expanded as decimal",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
+      maxInputTokens: {
+        value: 1e6,
+        source: "official-doc",
+        sourceRef: "https://cloud.tencent.com/document/product/1823/130051 — glm-5.3 row lists maximum input 1M tokens; k/M expanded as decimal",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
+      maxOutputTokens: {
+        value: 128000,
+        source: "official-doc",
+        sourceRef: "https://cloud.tencent.com/document/product/1823/130051 — glm-5.3 row lists maximum output 128k tokens; k/M expanded as decimal",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
       inputModalities: {
         value: [
           "text"
@@ -66660,6 +68247,27 @@ var catalog_default = {
       endpoints: [
         "chat"
       ],
+      contextWindow: {
+        value: 1e6,
+        source: "official-doc",
+        sourceRef: "https://cloud.tencent.com/document/product/1823/130051 — glm-5.3-flash row lists context window 1M tokens; k/M expanded as decimal",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
+      maxInputTokens: {
+        value: 1e6,
+        source: "official-doc",
+        sourceRef: "https://cloud.tencent.com/document/product/1823/130051 — glm-5.3-flash row lists maximum input 1M tokens; k/M expanded as decimal",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
+      maxOutputTokens: {
+        value: 128000,
+        source: "official-doc",
+        sourceRef: "https://cloud.tencent.com/document/product/1823/130051 — glm-5.3-flash row lists maximum output 128k tokens; k/M expanded as decimal",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
       inputModalities: {
         value: [
           "text"
@@ -66904,6 +68512,27 @@ var catalog_default = {
       endpoints: [
         "chat"
       ],
+      contextWindow: {
+        value: 256000,
+        source: "official-doc",
+        sourceRef: "https://cloud.tencent.com/document/product/1823/130051 — kimi-k2.7-code row lists context window 256k tokens; k/M expanded as decimal",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
+      maxInputTokens: {
+        value: 256000,
+        source: "official-doc",
+        sourceRef: "https://cloud.tencent.com/document/product/1823/130051 — kimi-k2.7-code row lists maximum input 256k tokens; k/M expanded as decimal",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
+      maxOutputTokens: {
+        value: 256000,
+        source: "official-doc",
+        sourceRef: "https://cloud.tencent.com/document/product/1823/130051 — kimi-k2.7-code row lists maximum output 256k tokens; k/M expanded as decimal",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
       inputModalities: {
         value: [
           "text"
@@ -66961,6 +68590,27 @@ var catalog_default = {
       endpoints: [
         "chat"
       ],
+      contextWindow: {
+        value: 1e6,
+        source: "official-doc",
+        sourceRef: "https://cloud.tencent.com/document/product/1823/130051 — kimi-k3 row lists context window 1M tokens; k/M expanded as decimal",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
+      maxInputTokens: {
+        value: 1e6,
+        source: "official-doc",
+        sourceRef: "https://cloud.tencent.com/document/product/1823/130051 — kimi-k3 row lists maximum input 1M tokens; k/M expanded as decimal",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
+      maxOutputTokens: {
+        value: 1e6,
+        source: "official-doc",
+        sourceRef: "https://cloud.tencent.com/document/product/1823/130051 — kimi-k3 row lists maximum output 1M tokens; k/M expanded as decimal",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
       inputModalities: {
         value: [
           "text"
@@ -67020,6 +68670,27 @@ var catalog_default = {
       endpoints: [
         "chat"
       ],
+      contextWindow: {
+        value: 200000,
+        source: "official-doc",
+        sourceRef: "https://cloud.tencent.com/document/product/1823/130051 — minimax-m2.7 row lists context window 200k tokens; k/M expanded as decimal",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
+      maxInputTokens: {
+        value: 200000,
+        source: "official-doc",
+        sourceRef: "https://cloud.tencent.com/document/product/1823/130051 — minimax-m2.7 row lists maximum input 200k tokens; k/M expanded as decimal",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
+      maxOutputTokens: {
+        value: 128000,
+        source: "official-doc",
+        sourceRef: "https://cloud.tencent.com/document/product/1823/130051 — minimax-m2.7 row lists maximum output 128k tokens; k/M expanded as decimal",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
       inputModalities: {
         value: [
           "text"
@@ -67079,6 +68750,20 @@ var catalog_default = {
       endpoints: [
         "chat"
       ],
+      contextWindow: {
+        value: 1e6,
+        source: "official-doc",
+        sourceRef: "https://cloud.tencent.com/document/product/1823/130051 — minimax-m3 row lists context window 1M tokens; k/M expanded as decimal",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
+      maxInputTokens: {
+        value: 1e6,
+        source: "official-doc",
+        sourceRef: "https://cloud.tencent.com/document/product/1823/130051 — minimax-m3 row lists maximum input 1M tokens; k/M expanded as decimal",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
       inputModalities: {
         value: [
           "text"
@@ -67185,6 +68870,27 @@ var catalog_default = {
       endpoints: [
         "chat"
       ],
+      contextWindow: {
+        value: 1e6,
+        source: "official-doc",
+        sourceRef: "https://cloud.tencent.com/document/product/1823/130051 — deepseek-v4-flash-202605 row lists context window 1M tokens; k/M expanded as decimal",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
+      maxInputTokens: {
+        value: 1e6,
+        source: "official-doc",
+        sourceRef: "https://cloud.tencent.com/document/product/1823/130051 — deepseek-v4-flash-202605 row lists maximum input 1M tokens; k/M expanded as decimal",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
+      maxOutputTokens: {
+        value: 384000,
+        source: "official-doc",
+        sourceRef: "https://cloud.tencent.com/document/product/1823/130051 — deepseek-v4-flash-202605 row lists maximum output 384k tokens; k/M expanded as decimal",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
       inputModalities: {
         value: [
           "text"
@@ -67249,6 +68955,27 @@ var catalog_default = {
       endpoints: [
         "chat"
       ],
+      contextWindow: {
+        value: 1e6,
+        source: "official-doc",
+        sourceRef: "https://cloud.tencent.com/document/product/1823/130051 — deepseek-v4-pro-202606 row lists context window 1M tokens; k/M expanded as decimal",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
+      maxInputTokens: {
+        value: 1e6,
+        source: "official-doc",
+        sourceRef: "https://cloud.tencent.com/document/product/1823/130051 — deepseek-v4-pro-202606 row lists maximum input 1M tokens; k/M expanded as decimal",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
+      maxOutputTokens: {
+        value: 384000,
+        source: "official-doc",
+        sourceRef: "https://cloud.tencent.com/document/product/1823/130051 — deepseek-v4-pro-202606 row lists maximum output 384k tokens; k/M expanded as decimal",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
       inputModalities: {
         value: [
           "text"
@@ -67312,6 +69039,27 @@ var catalog_default = {
       endpoints: [
         "chat"
       ],
+      contextWindow: {
+        value: 200000,
+        source: "official-doc",
+        sourceRef: "https://cloud.tencent.com/document/product/1823/130051 — glm-5 row lists context window 200k tokens; k/M expanded as decimal",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
+      maxInputTokens: {
+        value: 200000,
+        source: "official-doc",
+        sourceRef: "https://cloud.tencent.com/document/product/1823/130051 — glm-5 row lists maximum input 200k tokens; k/M expanded as decimal",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
+      maxOutputTokens: {
+        value: 128000,
+        source: "official-doc",
+        sourceRef: "https://cloud.tencent.com/document/product/1823/130051 — glm-5 row lists maximum output 128k tokens; k/M expanded as decimal",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
       inputModalities: {
         value: [
           "text"
@@ -67375,6 +69123,27 @@ var catalog_default = {
       endpoints: [
         "chat"
       ],
+      contextWindow: {
+        value: 200000,
+        source: "official-doc",
+        sourceRef: "https://cloud.tencent.com/document/product/1823/130051 — glm-5.1 row lists context window 200k tokens; k/M expanded as decimal",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
+      maxInputTokens: {
+        value: 200000,
+        source: "official-doc",
+        sourceRef: "https://cloud.tencent.com/document/product/1823/130051 — glm-5.1 row lists maximum input 200k tokens; k/M expanded as decimal",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
+      maxOutputTokens: {
+        value: 128000,
+        source: "official-doc",
+        sourceRef: "https://cloud.tencent.com/document/product/1823/130051 — glm-5.1 row lists maximum output 128k tokens; k/M expanded as decimal",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
       inputModalities: {
         value: [
           "text"
@@ -67438,6 +69207,27 @@ var catalog_default = {
       endpoints: [
         "chat"
       ],
+      contextWindow: {
+        value: 1e6,
+        source: "official-doc",
+        sourceRef: "https://cloud.tencent.com/document/product/1823/130051 — glm-5.2 row lists context window 1M tokens; k/M expanded as decimal",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
+      maxInputTokens: {
+        value: 1e6,
+        source: "official-doc",
+        sourceRef: "https://cloud.tencent.com/document/product/1823/130051 — glm-5.2 row lists maximum input 1M tokens; k/M expanded as decimal",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
+      maxOutputTokens: {
+        value: 128000,
+        source: "official-doc",
+        sourceRef: "https://cloud.tencent.com/document/product/1823/130051 — glm-5.2 row lists maximum output 128k tokens; k/M expanded as decimal",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
       inputModalities: {
         value: [
           "text"
@@ -67501,6 +69291,27 @@ var catalog_default = {
       endpoints: [
         "chat"
       ],
+      contextWindow: {
+        value: 1e6,
+        source: "official-doc",
+        sourceRef: "https://cloud.tencent.com/document/product/1823/130051 — glm-5.3 row lists context window 1M tokens; k/M expanded as decimal",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
+      maxInputTokens: {
+        value: 1e6,
+        source: "official-doc",
+        sourceRef: "https://cloud.tencent.com/document/product/1823/130051 — glm-5.3 row lists maximum input 1M tokens; k/M expanded as decimal",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
+      maxOutputTokens: {
+        value: 128000,
+        source: "official-doc",
+        sourceRef: "https://cloud.tencent.com/document/product/1823/130051 — glm-5.3 row lists maximum output 128k tokens; k/M expanded as decimal",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
       inputModalities: {
         value: [
           "text"
@@ -67562,6 +69373,27 @@ var catalog_default = {
       endpoints: [
         "chat"
       ],
+      contextWindow: {
+        value: 1e6,
+        source: "official-doc",
+        sourceRef: "https://cloud.tencent.com/document/product/1823/130051 — glm-5.3-flash row lists context window 1M tokens; k/M expanded as decimal",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
+      maxInputTokens: {
+        value: 1e6,
+        source: "official-doc",
+        sourceRef: "https://cloud.tencent.com/document/product/1823/130051 — glm-5.3-flash row lists maximum input 1M tokens; k/M expanded as decimal",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
+      maxOutputTokens: {
+        value: 128000,
+        source: "official-doc",
+        sourceRef: "https://cloud.tencent.com/document/product/1823/130051 — glm-5.3-flash row lists maximum output 128k tokens; k/M expanded as decimal",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
       inputModalities: {
         value: [
           "text"
@@ -67818,6 +69650,27 @@ var catalog_default = {
       endpoints: [
         "chat"
       ],
+      contextWindow: {
+        value: 256000,
+        source: "official-doc",
+        sourceRef: "https://cloud.tencent.com/document/product/1823/130051 — kimi-k2.7-code row lists context window 256k tokens; k/M expanded as decimal",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
+      maxInputTokens: {
+        value: 256000,
+        source: "official-doc",
+        sourceRef: "https://cloud.tencent.com/document/product/1823/130051 — kimi-k2.7-code row lists maximum input 256k tokens; k/M expanded as decimal",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
+      maxOutputTokens: {
+        value: 256000,
+        source: "official-doc",
+        sourceRef: "https://cloud.tencent.com/document/product/1823/130051 — kimi-k2.7-code row lists maximum output 256k tokens; k/M expanded as decimal",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
       inputModalities: {
         value: [
           "text"
@@ -67879,6 +69732,27 @@ var catalog_default = {
       endpoints: [
         "chat"
       ],
+      contextWindow: {
+        value: 1e6,
+        source: "official-doc",
+        sourceRef: "https://cloud.tencent.com/document/product/1823/130051 — kimi-k3 row lists context window 1M tokens; k/M expanded as decimal",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
+      maxInputTokens: {
+        value: 1e6,
+        source: "official-doc",
+        sourceRef: "https://cloud.tencent.com/document/product/1823/130051 — kimi-k3 row lists maximum input 1M tokens; k/M expanded as decimal",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
+      maxOutputTokens: {
+        value: 1e6,
+        source: "official-doc",
+        sourceRef: "https://cloud.tencent.com/document/product/1823/130051 — kimi-k3 row lists maximum output 1M tokens; k/M expanded as decimal",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
       inputModalities: {
         value: [
           "text"
@@ -67942,6 +69816,27 @@ var catalog_default = {
       endpoints: [
         "chat"
       ],
+      contextWindow: {
+        value: 200000,
+        source: "official-doc",
+        sourceRef: "https://cloud.tencent.com/document/product/1823/130051 — minimax-m2.7 row lists context window 200k tokens; k/M expanded as decimal",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
+      maxInputTokens: {
+        value: 200000,
+        source: "official-doc",
+        sourceRef: "https://cloud.tencent.com/document/product/1823/130051 — minimax-m2.7 row lists maximum input 200k tokens; k/M expanded as decimal",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
+      maxOutputTokens: {
+        value: 128000,
+        source: "official-doc",
+        sourceRef: "https://cloud.tencent.com/document/product/1823/130051 — minimax-m2.7 row lists maximum output 128k tokens; k/M expanded as decimal",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
       inputModalities: {
         value: [
           "text"
@@ -68005,6 +69900,20 @@ var catalog_default = {
       endpoints: [
         "chat"
       ],
+      contextWindow: {
+        value: 1e6,
+        source: "official-doc",
+        sourceRef: "https://cloud.tencent.com/document/product/1823/130051 — minimax-m3 row lists context window 1M tokens; k/M expanded as decimal",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
+      maxInputTokens: {
+        value: 1e6,
+        source: "official-doc",
+        sourceRef: "https://cloud.tencent.com/document/product/1823/130051 — minimax-m3 row lists maximum input 1M tokens; k/M expanded as decimal",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
       inputModalities: {
         value: [
           "text"
@@ -68552,14 +70461,14 @@ var catalog_default = {
       },
       pricing: {
         value: {
-          inputPerMTokUsd: 0.149,
-          outputPerMTokUsd: 0.596,
-          cacheReadPerMTokUsd: 0.03725
+          inputPerMTokUsd: 0.132,
+          outputPerMTokUsd: 0.528,
+          cacheReadPerMTokUsd: 0.033
         },
         source: "official-doc",
-        sourceRef: "https://cloud.tencent.com/document/product/1823/130055 — hy3: CNY 1 input, 4 output, 0.25 cache hit per million tokens. Converted at 2026-09-25 1 CNY = $0.1490 (https://www.investing.com/currencies/cny-usd-historical-data).",
-        confidence: "inferred",
-        observedAt: "2026-09-25T10:06:30Z"
+        sourceRef: "https://proxy-hk.tencentcloud.com/pt/document/product/1300/78937 — Singapore international USD list, hy3: $0.132 input/$0.528 output/$0.033 cache-hit per million tokens",
+        confidence: "declared",
+        observedAt: "2026-09-25T12:31:56.381Z"
       },
       unsupportedParameters: [],
       status: "candidate",
@@ -68655,14 +70564,14 @@ var catalog_default = {
       },
       pricing: {
         value: {
-          inputPerMTokUsd: 0.894,
-          outputPerMTokUsd: 2.682,
-          cacheReadPerMTokUsd: 0.0447
+          inputPerMTokUsd: 0.834,
+          outputPerMTokUsd: 2.501,
+          cacheReadPerMTokUsd: 0.042
         },
         source: "official-doc",
-        sourceRef: "https://cloud.tencent.com/document/product/1823/130055 — hy4-preview: CNY 6 input, 18 output, 0.3 cache hit per million tokens. Converted at 2026-09-25 1 CNY = $0.1490 (https://www.investing.com/currencies/cny-usd-historical-data).",
-        confidence: "inferred",
-        observedAt: "2026-09-25T10:06:30Z"
+        sourceRef: "https://proxy-hk.tencentcloud.com/pt/document/product/1300/78937 — Singapore international USD list, hy4-preview: $0.834 input/$2.501 output/$0.042 cache-hit per million tokens",
+        confidence: "declared",
+        observedAt: "2026-09-25T12:31:56.381Z"
       },
       unsupportedParameters: [],
       status: "candidate",
@@ -68762,14 +70671,14 @@ var catalog_default = {
       },
       pricing: {
         value: {
-          inputPerMTokUsd: 0.149,
-          outputPerMTokUsd: 0.596,
-          cacheReadPerMTokUsd: 0.03725
+          inputPerMTokUsd: 0.132,
+          outputPerMTokUsd: 0.528,
+          cacheReadPerMTokUsd: 0.033
         },
         source: "official-doc",
-        sourceRef: "https://cloud.tencent.com/document/product/1823/130055 — hy3: CNY 1 input, 4 output, 0.25 cache hit per million tokens. Converted at 2026-09-25 1 CNY = $0.1490 (https://www.investing.com/currencies/cny-usd-historical-data).",
-        confidence: "inferred",
-        observedAt: "2026-09-25T10:06:30Z"
+        sourceRef: "https://proxy-hk.tencentcloud.com/pt/document/product/1300/78937 — Singapore international USD list, hy3: $0.132 input/$0.528 output/$0.033 cache-hit per million tokens",
+        confidence: "declared",
+        observedAt: "2026-09-25T12:31:56.381Z"
       },
       unsupportedParameters: [],
       status: "candidate",
@@ -68869,14 +70778,14 @@ var catalog_default = {
       },
       pricing: {
         value: {
-          inputPerMTokUsd: 0.894,
-          outputPerMTokUsd: 2.682,
-          cacheReadPerMTokUsd: 0.0447
+          inputPerMTokUsd: 0.834,
+          outputPerMTokUsd: 2.501,
+          cacheReadPerMTokUsd: 0.042
         },
         source: "official-doc",
-        sourceRef: "https://cloud.tencent.com/document/product/1823/130055 — hy4-preview: CNY 6 input, 18 output, 0.3 cache hit per million tokens. Converted at 2026-09-25 1 CNY = $0.1490 (https://www.investing.com/currencies/cny-usd-historical-data).",
-        confidence: "inferred",
-        observedAt: "2026-09-25T10:06:30Z"
+        sourceRef: "https://proxy-hk.tencentcloud.com/pt/document/product/1300/78937 — Singapore international USD list, hy4-preview: $0.834 input/$2.501 output/$0.042 cache-hit per million tokens",
+        confidence: "declared",
+        observedAt: "2026-09-25T12:31:56.381Z"
       },
       unsupportedParameters: [],
       status: "candidate",
@@ -72973,7 +74882,8 @@ var catalog_default = {
         "grok-4.20-non-reasoning-gv2"
       ],
       endpoints: [
-        "chat"
+        "chat",
+        "responses"
       ],
       contextWindow: {
         value: 1e6,
@@ -73068,7 +74978,8 @@ var catalog_default = {
         "grok-4.20-reasoning-gv2"
       ],
       endpoints: [
-        "chat"
+        "chat",
+        "responses"
       ],
       contextWindow: {
         value: 1e6,
@@ -73133,7 +75044,7 @@ var catalog_default = {
           observedAt: "2026-09-25T08:40:00Z"
         },
         efforts: [],
-        continuation: "none"
+        continuation: "opaque-provider-state"
       },
       pricing: {
         value: {
@@ -73151,6 +75062,111 @@ var catalog_default = {
       canonicalModelId: "grok-4.20-0309-reasoning",
       modelFamily: "grok"
     },
+    {
+      key: "xai/grok-4.20-multi-agent-0309",
+      providerId: "xai",
+      upstreamId: "grok-4.20-multi-agent-0309",
+      displayName: "Grok 4.20 Multi-Agent Beta",
+      aliases: [
+        "grok-4.20-multi-agent",
+        "grok-4.20-multi-agent-latest",
+        "grok-4.20-multi-agent-beta-latest",
+        "grok-4.20-multi-agent-experimental-beta-0304",
+        "grok-4.20-multi-agent-experimental-beta-latest",
+        "grok-4.20-multi-agent-beta-0309"
+      ],
+      endpoints: [
+        "responses"
+      ],
+      contextWindow: {
+        value: 1e6,
+        source: "official-doc",
+        sourceRef: "https://docs.x.ai/developers/models/grok-4.20-multi-agent-0309 — model page context window 1,000,000 tokens",
+        confidence: "declared",
+        observedAt: "2026-09-25T08:40:00Z"
+      },
+      inputModalities: {
+        value: [
+          "text",
+          "image"
+        ],
+        source: "official-doc",
+        sourceRef: "https://docs.x.ai/developers/models/grok-4.20-multi-agent-0309 — model page Modalities: Text, Image input",
+        confidence: "declared",
+        observedAt: "2026-09-25T08:40:00Z"
+      },
+      outputModalities: {
+        value: [
+          "text"
+        ],
+        source: "official-doc",
+        sourceRef: "https://docs.x.ai/developers/models/grok-4.20-multi-agent-0309 — model page Modalities: Text output",
+        confidence: "declared",
+        observedAt: "2026-09-25T08:40:00Z"
+      },
+      toolCalling: {
+        value: "none",
+        source: "official-doc",
+        sourceRef: "https://docs.x.ai/developers/model-capabilities/text/multi-agent — multi-agent limitations: client-side/custom function calling unsupported; built-in server tools only",
+        confidence: "declared",
+        observedAt: "2026-09-25T08:40:00Z"
+      },
+      nativeTools: {
+        value: false,
+        source: "official-doc",
+        sourceRef: "https://docs.x.ai/developers/model-capabilities/text/multi-agent — client-side/custom function tools unsupported",
+        confidence: "declared",
+        observedAt: "2026-09-25T08:40:00Z"
+      },
+      structuredOutput: {
+        value: true,
+        source: "official-doc",
+        sourceRef: "https://docs.x.ai/developers/models/grok-4.20-multi-agent-0309 — model page Structured outputs capability",
+        confidence: "declared",
+        observedAt: "2026-09-25T08:40:00Z"
+      },
+      promptCaching: {
+        value: true,
+        source: "official-doc",
+        sourceRef: "https://docs.x.ai/developers/models/grok-4.20-multi-agent-0309 — model page lists Cached tokens input rate; prompt caching available",
+        confidence: "declared",
+        observedAt: "2026-09-25T08:40:00Z"
+      },
+      reasoning: {
+        supported: {
+          value: true,
+          source: "official-doc",
+          sourceRef: "https://docs.x.ai/developers/model-capabilities/text/multi-agent — reasoning.effort low/medium selects 4 agents, high/xhigh selects 16; previous_response_id supports multi-turn",
+          confidence: "declared",
+          observedAt: "2026-09-25T08:40:00Z"
+        },
+        efforts: [
+          "low",
+          "medium",
+          "high",
+          "xhigh"
+        ],
+        continuation: "opaque-provider-state"
+      },
+      pricing: {
+        value: {
+          inputPerMTokUsd: 1.25,
+          outputPerMTokUsd: 2.5,
+          cacheReadPerMTokUsd: 0.2
+        },
+        source: "official-doc",
+        sourceRef: "https://docs.x.ai/developers/pricing — grok-4.20-multi-agent-0309 Standard global short-context rate (<200k prompt tokens): $1.25 input, $0.20 cached input, $2.50 output per 1M; >=200k the entire request is charged $2.50/$0.40/$5.00. US-regional 1.1x applies only to models currently on that endpoint, listed as grok-4.7 and grok-4.6; not this model (re-read 2026-09-25, WS-23).",
+        confidence: "declared",
+        observedAt: "2026-09-25T17:09:48Z"
+      },
+      unsupportedParameters: [
+        "max_output_tokens",
+        "tools"
+      ],
+      status: "candidate",
+      canonicalModelId: "grok-4.20-multi-agent-0309",
+      modelFamily: "grok"
+    },
     {
       key: "xai/grok-4.3",
       providerId: "xai",
@@ -73160,7 +75176,8 @@ var catalog_default = {
         "grok-4.3-latest"
       ],
       endpoints: [
-        "chat"
+        "chat",
+        "responses"
       ],
       contextWindow: {
         value: 1e6,
@@ -73231,7 +75248,7 @@ var catalog_default = {
           "high",
           "xhigh"
         ],
-        continuation: "plaintext",
+        continuation: "opaque-provider-state",
         defaultEffort: "low"
       },
       pricing: {
@@ -73241,9 +75258,9 @@ var catalog_default = {
           cacheReadPerMTokUsd: 0.2
         },
         source: "official-doc",
-        sourceRef: "https://docs.x.ai/developers/pricing — current global Standard Text API price table for grok-4.3; it charges the listed long-context rate for every token after the prompt reaches 200K and applies a 10% US-regional premium. This catalog records the standard short-context rate (retrieved 2026-09-19).",
+        sourceRef: "https://docs.x.ai/developers/pricing — grok-4.3 Standard global short-context rate: $1.25 input/$2.5 output/$0.2 cached per million; long-context threshold 200K. US-regional 1.1x applies only to models currently on that endpoint, listed as grok-4.7 and grok-4.6; not grok-4.3.",
         confidence: "declared",
-        observedAt: "2026-09-19T00:00:00Z"
+        observedAt: "2026-09-25T12:31:56.381Z"
       },
       unsupportedParameters: [],
       status: "candidate",
@@ -73260,7 +75277,8 @@ var catalog_default = {
         "grok-build-latest"
       ],
       endpoints: [
-        "chat"
+        "chat",
+        "responses"
       ],
       contextWindow: {
         value: 500000,
@@ -73329,7 +75347,7 @@ var catalog_default = {
           "medium",
           "high"
         ],
-        continuation: "plaintext",
+        continuation: "opaque-provider-state",
         defaultEffort: "high"
       },
       pricing: {
@@ -73359,7 +75377,8 @@ var catalog_default = {
       displayName: "Grok 4.6",
       aliases: [],
       endpoints: [
-        "chat"
+        "chat",
+        "responses"
       ],
       contextWindow: {
         value: 500000,
@@ -73429,7 +75448,7 @@ var catalog_default = {
           "high",
           "xhigh"
         ],
-        continuation: "plaintext",
+        continuation: "opaque-provider-state",
         defaultEffort: "high"
       },
       pricing: {
@@ -73459,7 +75478,8 @@ var catalog_default = {
       displayName: "Grok 4.7",
       aliases: [],
       endpoints: [
-        "chat"
+        "chat",
+        "responses"
       ],
       contextWindow: {
         value: 500000,
@@ -73529,8 +75549,29 @@ var catalog_default = {
           "high",
           "xhigh"
         ],
-        continuation: "plaintext",
-        defaultEffort: "high"
+        continuation: "opaque-provider-state",
+        defaultEffort: "high",
+        readableState: {
+          value: "summary",
+          source: "official-doc",
+          sourceRef: 'https://docs.x.ai/developers/model-capabilities/text/reasoning — "For `grok-4.7`, we expose summarizations of the model\'s internal reasoning" (Summarized Reasoning Content); https://docs.x.ai/developers/rest-api-reference/inference/responses — the example reasoning output item carries `summary: [{ type: "summary_text", … }]`',
+          confidence: "declared",
+          observedAt: "2026-09-25T17:09:48Z"
+        },
+        summaryRequest: {
+          value: {
+            field: "reasoning.summary",
+            values: [
+              "detailed",
+              "auto",
+              "concise"
+            ]
+          },
+          source: "official-doc",
+          sourceRef: 'https://docs.x.ai/developers/rest-api-reference/inference/responses — `reasoning.summary`: "Possible values are `auto`, `concise` and `detailed`. Only included for compatibility. The model shall always return `detailed`." `detailed` is listed FIRST because the adapter sends the first value, and it is the one the model returns regardless',
+          confidence: "declared",
+          observedAt: "2026-09-25T17:09:48Z"
+        }
       },
       pricing: {
         value: {
@@ -73563,7 +75604,8 @@ var catalog_default = {
         "grok-code-fast-1-0825"
       ],
       endpoints: [
-        "chat"
+        "chat",
+        "responses"
       ],
       contextWindow: {
         value: 256000,
@@ -73628,7 +75670,7 @@ var catalog_default = {
           observedAt: "2026-09-25T08:40:00Z"
         },
         efforts: [],
-        continuation: "none"
+        continuation: "opaque-provider-state"
       },
       pricing: {
         value: {
@@ -73637,9 +75679,9 @@ var catalog_default = {
           cacheReadPerMTokUsd: 0.2
         },
         source: "official-doc",
-        sourceRef: "https://docs.x.ai/developers/pricing — current global Standard Text API price table for grok-build-0.1; it charges the listed long-context rate for every token after the prompt reaches 200K and applies a 10% US-regional premium. This catalog records the standard short-context rate (retrieved 2026-09-19).",
+        sourceRef: "https://docs.x.ai/developers/pricing — grok-build-0.1 Standard global short-context rate: $1 input/$2 output/$0.2 cached per million; long-context threshold 200K. US-regional 1.1x applies only to models currently on that endpoint, listed as grok-4.7 and grok-4.6; not grok-build-0.1.",
         confidence: "declared",
-        observedAt: "2026-09-19T00:00:00Z"
+        observedAt: "2026-09-25T12:31:56.381Z"
       },
       unsupportedParameters: [],
       status: "candidate",
@@ -73741,7 +75783,7 @@ var catalog_default = {
         observedAt: "2026-09-25T09:52:37Z"
       },
       unsupportedParameters: [],
-      status: "deprecated",
+      status: "candidate",
       canonicalModelId: "mimo-v2.5",
       modelFamily: "mimo"
     },
@@ -73837,7 +75879,7 @@ var catalog_default = {
         observedAt: "2026-09-25T09:52:37Z"
       },
       unsupportedParameters: [],
-      status: "deprecated",
+      status: "candidate",
       canonicalModelId: "mimo-v2.5-pro",
       modelFamily: "mimo"
     },
@@ -74222,7 +76264,7 @@ var catalog_default = {
         continuation: "none"
       },
       unsupportedParameters: [],
-      status: "deprecated",
+      status: "candidate",
       canonicalModelId: "mimo-v2.5",
       modelFamily: "mimo"
     },
@@ -74307,7 +76349,7 @@ var catalog_default = {
         continuation: "none"
       },
       unsupportedParameters: [],
-      status: "deprecated",
+      status: "candidate",
       canonicalModelId: "mimo-v2.5-pro",
       modelFamily: "mimo"
     },
@@ -74571,7 +76613,7 @@ var catalog_default = {
         continuation: "none"
       },
       unsupportedParameters: [],
-      status: "deprecated",
+      status: "candidate",
       canonicalModelId: "mimo-v2.5",
       modelFamily: "mimo"
     },
@@ -74656,7 +76698,7 @@ var catalog_default = {
         continuation: "none"
       },
       unsupportedParameters: [],
-      status: "deprecated",
+      status: "candidate",
       canonicalModelId: "mimo-v2.5-pro",
       modelFamily: "mimo"
     },
@@ -74920,7 +76962,7 @@ var catalog_default = {
         continuation: "plaintext"
       },
       unsupportedParameters: [],
-      status: "deprecated",
+      status: "candidate",
       canonicalModelId: "mimo-v2.5",
       modelFamily: "mimo"
     },
@@ -75005,7 +77047,7 @@ var catalog_default = {
         continuation: "plaintext"
       },
       unsupportedParameters: [],
-      status: "deprecated",
+      status: "candidate",
       canonicalModelId: "mimo-v2.5-pro",
       modelFamily: "mimo"
     },
@@ -75269,7 +77311,7 @@ var catalog_default = {
         continuation: "none"
       },
       unsupportedParameters: [],
-      status: "deprecated",
+      status: "candidate",
       canonicalModelId: "mimo-v2.5",
       modelFamily: "mimo"
     },
@@ -75354,7 +77396,7 @@ var catalog_default = {
         continuation: "none"
       },
       unsupportedParameters: [],
-      status: "deprecated",
+      status: "candidate",
       canonicalModelId: "mimo-v2.5-pro",
       modelFamily: "mimo"
     },
@@ -75618,7 +77660,7 @@ var catalog_default = {
         continuation: "plaintext"
       },
       unsupportedParameters: [],
-      status: "deprecated",
+      status: "candidate",
       canonicalModelId: "mimo-v2.5",
       modelFamily: "mimo"
     },
@@ -75703,7 +77745,7 @@ var catalog_default = {
         continuation: "plaintext"
       },
       unsupportedParameters: [],
-      status: "deprecated",
+      status: "candidate",
       canonicalModelId: "mimo-v2.5-pro",
       modelFamily: "mimo"
     },
@@ -75967,7 +78009,7 @@ var catalog_default = {
         continuation: "plaintext"
       },
       unsupportedParameters: [],
-      status: "deprecated",
+      status: "candidate",
       canonicalModelId: "mimo-v2.5",
       modelFamily: "mimo"
     },
@@ -76052,7 +78094,7 @@ var catalog_default = {
         continuation: "plaintext"
       },
       unsupportedParameters: [],
-      status: "deprecated",
+      status: "candidate",
       canonicalModelId: "mimo-v2.5-pro",
       modelFamily: "mimo"
     },
@@ -76327,7 +78369,7 @@ var catalog_default = {
         observedAt: "2026-09-25T09:52:37Z"
       },
       unsupportedParameters: [],
-      status: "deprecated",
+      status: "candidate",
       canonicalModelId: "mimo-v2.5",
       modelFamily: "mimo"
     },
@@ -76423,7 +78465,7 @@ var catalog_default = {
         observedAt: "2026-09-25T09:52:37Z"
       },
       unsupportedParameters: [],
-      status: "deprecated",
+      status: "candidate",
       canonicalModelId: "mimo-v2.5-pro",
       modelFamily: "mimo"
     },
@@ -77691,6 +79733,13 @@ var catalog_default = {
       endpoints: [
         "chat"
       ],
+      contextWindow: {
+        value: 128000,
+        source: "official-doc",
+        sourceRef: "https://docs.z.ai/guides/llm/glm-4.5 — glm-4.5 model guide/table gives 128K context length; K expanded as 1,000 tokens",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
       maxOutputTokens: {
         value: 98304,
         source: "official-doc",
@@ -77773,6 +79822,13 @@ var catalog_default = {
       endpoints: [
         "chat"
       ],
+      contextWindow: {
+        value: 128000,
+        source: "official-doc",
+        sourceRef: "https://docs.z.ai/llms-full.txt — glm-4.5-air model guide/table gives 128K context length; K expanded as 1,000 tokens",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
       maxOutputTokens: {
         value: 98304,
         source: "official-doc",
@@ -77855,6 +79911,13 @@ var catalog_default = {
       endpoints: [
         "chat"
       ],
+      contextWindow: {
+        value: 200000,
+        source: "official-doc",
+        sourceRef: "https://docs.z.ai/guides/llm/glm-4.6 — glm-4.6 model guide/table gives 200K context length; K expanded as 1,000 tokens",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
       maxOutputTokens: {
         value: 131072,
         source: "official-doc",
@@ -77937,6 +80000,13 @@ var catalog_default = {
       endpoints: [
         "chat"
       ],
+      contextWindow: {
+        value: 200000,
+        source: "official-doc",
+        sourceRef: "https://docs.z.ai/guides/llm/glm-4.7 — glm-4.7 model guide/table gives 200K context length; K expanded as 1,000 tokens",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
       maxOutputTokens: {
         value: 131072,
         source: "official-doc",
@@ -78040,6 +80110,13 @@ var catalog_default = {
       endpoints: [
         "chat"
       ],
+      contextWindow: {
+        value: 200000,
+        source: "official-doc",
+        sourceRef: "https://docs.z.ai/llms-full.txt — glm-4.7-flash model guide/table gives 200K context length; K expanded as 1,000 tokens",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
       inputModalities: {
         value: [
           "text"
@@ -78129,6 +80206,13 @@ var catalog_default = {
       endpoints: [
         "chat"
       ],
+      contextWindow: {
+        value: 200000,
+        source: "official-doc",
+        sourceRef: "https://docs.z.ai/llms-full.txt — glm-4.7-flashx model guide/table gives 200K context length; K expanded as 1,000 tokens",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
       inputModalities: {
         value: [
           "text"
@@ -78204,6 +80288,13 @@ var catalog_default = {
       endpoints: [
         "chat"
       ],
+      contextWindow: {
+        value: 200000,
+        source: "official-doc",
+        sourceRef: "https://docs.z.ai/guides/llm/glm-5 — glm-5 model guide/table gives 200K context length; K expanded as 1,000 tokens",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
       maxOutputTokens: {
         value: 131072,
         source: "official-doc",
@@ -78307,6 +80398,13 @@ var catalog_default = {
       endpoints: [
         "chat"
       ],
+      contextWindow: {
+        value: 200000,
+        source: "official-doc",
+        sourceRef: "https://docs.z.ai/guides/llm/glm-5-turbo — glm-5-turbo model guide/table gives 200K context length; K expanded as 1,000 tokens",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
       inputModalities: {
         value: [
           "text"
@@ -78385,6 +80483,13 @@ var catalog_default = {
       endpoints: [
         "chat"
       ],
+      contextWindow: {
+        value: 200000,
+        source: "official-doc",
+        sourceRef: "https://docs.z.ai/guides/llm/glm-5.1 — glm-5.1 model guide/table gives 200K context length; K expanded as 1,000 tokens",
+        confidence: "inferred",
+        observedAt: "2026-09-25T12:31:56.381Z"
+      },
       maxOutputTokens: {
         value: 131072,
         source: "official-doc",