@oonacode/win32-x64 0.3.235 → 0.3.236

package/oona.exe

@@ -1,4 +1,4 @@
 [diffend] Oversized file quarantined before diffing.
 name: package/oona.exe
-size: 112560640 bytes
-sha256: e69227389f4f8adc8bd2ac11f4b0248066858ba84ef339cb193203dbf3241d06
+size: 112585728 bytes
+sha256: eab93c3ca7ab04cb0267c1b64fdc0e4bea7b7ef776ac1491b1aaee4156ebed57

package/oonacode-release.json

@@ -1,13 +1,13 @@
 {
     "name":  "oonacode",
-    "version":  "0.3.235",
+    "version":  "0.3.236",
     "channel":  "stable",
     "platform":  "win32",
     "arch":  "x64",
     "mode":  "bun",
     "launcherRuntime":  "bun 1.4.1",
-    "commit":  "eb6bcc8d",
-    "builtAt":  "2026-10-13T19:09:35.6378719Z",
+    "commit":  "1b402b08",
+    "builtAt":  "2026-10-13T23:00:54.9740290Z",
     "engine":  "engine-store",
     "engineVersion":  "0.3.296",
     "siteOrigin":  "https://oonacode.oonak.ai",

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@oonacode/win32-x64",
-  "version": "0.3.235",
+  "version": "0.3.236",
   "description": "OonaCode native binaries for Windows x64: oona.exe (CLI + headless host service), the pinned oonaclaude.exe engine sidecar, local web UI assets. Installed automatically as an optional dependency of `oonacode`.",
   "license": "UNLICENSED",
   "homepage": "https://oonacode.oonak.ai",

package/registry.json

@@ -116,7 +116,8 @@
           "supportsVision": true,
           "inputPricePerM": 4,
           "outputPricePerM": 20,
-          "notes": "The Opus after Opus 5, cheaper than it: $4 / $20 per MTok, cache reads $0.20, 1M context, 128K output (Anthropic's model table, 2026-09-23). Needs Claude Code 2.1.280 or newer: Anthropic answers any request carrying an older engine's identity with 400 'version 2.1.280 or newer is required', on the API key and the subscription alike (measured 2026-09-23 with engine 2.1.272), which is why it arrived with the move to Agent SDK 0.3.280. Thinking cannot be disabled and forced tool_choice is refused; its effort default is medium, one below Opus 5's high. Listed AFTER Opus 5 on purpose, on both Claude entries: neither declares modelAliases, so the 'opus' family (ANTHROPIC_DEFAULT_OPUS_MODEL, and the word 'opus' asked of the pool) resolves to the first listed model whose id says opus, and it stays on Opus 5, the default (2026-09-23 review). An old Opus ID the pool does not carry is refused pool_model_unknown since 2026-09-28, never answered by Opus 5."
+          "notes": "The Opus after Opus 5, cheaper than it: $4 / $20 per MTok, cache reads $0.20, 1M context, 128K output (Anthropic's model table, 2026-09-23). Needs Claude Code 2.1.280 or newer: Anthropic answers any request carrying an older engine's identity with 400 'version 2.1.280 or newer is required', on the API key and the subscription alike (measured 2026-09-23 with engine 2.1.272), which is why it arrived with the move to Agent SDK 0.3.280. Thinking cannot be disabled and forced tool_choice is refused; its effort default is medium, one below Opus 5's high. Listed AFTER Opus 5 on purpose, on both Claude entries: neither declares modelAliases, so the 'opus' family (ANTHROPIC_DEFAULT_OPUS_MODEL, and the word 'opus' asked of the pool) resolves to the first listed model whose id says opus, and it stays on Opus 5, the default (2026-09-23 review). An old Opus ID the pool does not carry is refused pool_model_unknown since 2026-09-28, never answered by Opus 5.",
+          "defaultEffort": "medium"
         },
         {
           "id": "claude-sonnet-5-5",
@@ -138,7 +139,8 @@
           "supportsVision": true,
           "inputPricePerM": 0.1,
           "outputPricePerM": 0.5,
-          "notes": "The Haiku after Haiku 4.5, released 2026-10-07, at $0.10 / $0.50 per MTok for prompts up to 100k, cache reads $0.01, 5-minute cache write $0.125; 1M context, 128K output (Anthropic's model page and price list, read 2026-10-08). It REPLACES Haiku 4.5 here: Haiku 4.5 is no longer listed, and its id is an admin alias of this row (migration 0179), so a chat or a program still naming claude-haiku-4-5 is served — and billed — as Haiku 5.5. Being the only Haiku listed, it is what the 'haiku' family word resolves to. Engine 2.1.280 does not know the id but serves it (probed from the production gateway 2026-10-08: subscription token 200, both with adaptive thinking and disabled thinking)."
+          "notes": "The Haiku after Haiku 4.5, released 2026-10-07, at $0.10 / $0.50 per MTok for prompts up to 100k, cache reads $0.01, 5-minute cache write $0.125; 1M context, 128K output (Anthropic's model page and price list, read 2026-10-08). It REPLACES Haiku 4.5 here: Haiku 4.5 is no longer listed, and its id is an admin alias of this row (migration 0179), so a chat or a program still naming claude-haiku-4-5 is served — and billed — as Haiku 5.5. Being the only Haiku listed, it is what the 'haiku' family word resolves to. Engine 2.1.280 does not know the id but serves it (probed from the production gateway 2026-10-08: subscription token 200, both with adaptive thinking and disabled thinking).",
+          "defaultEffort": "medium"
         }
       ],
       "defaultModel": "claude-opus-5",
@@ -189,7 +191,8 @@
           "operatorOnly": true
         }
       ],
-      "recordLabel": "API"
+      "recordLabel": "API",
+      "defaultEffort": "high"
     },
     {
       "id": "openai",
@@ -227,7 +230,8 @@
           "contextWindow": 1000000,
           "supportsTools": true,
           "supportsThinking": false,
-          "supportsVision": true
+          "supportsVision": true,
+          "effortLevels": []
         },
         {
           "id": "o4-mini",
@@ -263,7 +267,8 @@
       },
       "phase": 1,
       "enabled": true,
-      "docsUrl": "https://platform.openai.com/docs"
+      "docsUrl": "https://platform.openai.com/docs",
+      "defaultEffort": "medium"
     },
     {
       "id": "openai-subscription",
@@ -292,7 +297,8 @@
           "supportsThinking": true,
           "supportsVision": true,
           "effortLevels": ["low", "medium", "high", "xhigh", "max"],
-          "notes": "Codex's first choice (priority 4 in its model list). Served from the person's own ChatGPT plan through the Codex backend; no per-token price - the plan's 5-hour and weekly windows are what it costs."
+          "notes": "Codex's first choice (priority 4 in its model list). Served from the person's own ChatGPT plan through the Codex backend; no per-token price - the plan's 5-hour and weekly windows are what it costs.",
+          "defaultEffort": "medium"
         },
         {
           "id": "gpt-5.6-terra",
@@ -302,7 +308,8 @@
           "supportsThinking": true,
           "supportsVision": true,
           "effortLevels": ["low", "medium", "high", "xhigh", "max"],
-          "notes": "Served from the person's own ChatGPT plan through the Codex backend; no per-token price - the plan's 5-hour and weekly windows are what it costs."
+          "notes": "Served from the person's own ChatGPT plan through the Codex backend; no per-token price - the plan's 5-hour and weekly windows are what it costs.",
+          "defaultEffort": "medium"
         },
         {
           "id": "gpt-5.6-luna",
@@ -312,7 +319,8 @@
           "supportsThinking": true,
           "supportsVision": true,
           "effortLevels": ["low", "medium", "high", "xhigh", "max"],
-          "notes": "Served from the person's own ChatGPT plan through the Codex backend; no per-token price - the plan's 5-hour and weekly windows are what it costs."
+          "notes": "Served from the person's own ChatGPT plan through the Codex backend; no per-token price - the plan's 5-hour and weekly windows are what it costs.",
+          "defaultEffort": "medium"
         },
         {
           "id": "gpt-5.5",
@@ -322,7 +330,8 @@
           "supportsThinking": true,
           "supportsVision": true,
           "effortLevels": ["low", "medium", "high", "xhigh"],
-          "notes": "Legacy: Codex hides it from its picker but the backend still serves it, and older chats are on it. No max effort. Served from the person's own ChatGPT plan through the Codex backend; no per-token price - the plan's 5-hour and weekly windows are what it costs."
+          "notes": "Legacy: Codex hides it from its picker but the backend still serves it, and older chats are on it. No max effort. Served from the person's own ChatGPT plan through the Codex backend; no per-token price - the plan's 5-hour and weekly windows are what it costs.",
+          "defaultEffort": "medium"
         },
         {
           "id": "gpt-6.1-sol",
@@ -332,7 +341,8 @@
           "supportsThinking": true,
           "supportsVision": true,
           "effortLevels": ["low", "medium", "high", "xhigh", "max"],
-          "notes": "Listed by the Codex backend for the operator's ChatGPT account on 2026-10-13 and answered a real turn through the genuine Codex CLI. Served from the person's own ChatGPT plan through the Codex backend; no per-token price - the plan's 5-hour and weekly windows are what it costs. The Codex backend takes 500k of context. Codex ultra effort is refused by the backend."
+          "notes": "Listed by the Codex backend for the operator's ChatGPT account on 2026-10-13 and answered a real turn through the genuine Codex CLI. Served from the person's own ChatGPT plan through the Codex backend; no per-token price - the plan's 5-hour and weekly windows are what it costs. The Codex backend takes 500k of context. Codex ultra effort is refused by the backend.",
+          "defaultEffort": "low"
         },
         {
           "id": "gpt-6-sol",
@@ -342,7 +352,8 @@
           "supportsThinking": true,
           "supportsVision": true,
           "effortLevels": ["low", "medium", "high", "xhigh", "max"],
-          "notes": "Listed by the Codex backend for the operator's ChatGPT account on 2026-10-13 and answered a real turn through the genuine Codex CLI. Served from the person's own ChatGPT plan through the Codex backend; no per-token price - the plan's 5-hour and weekly windows are what it costs. The Codex backend takes 272k of context. Codex ultra effort is refused by the backend."
+          "notes": "Listed by the Codex backend for the operator's ChatGPT account on 2026-10-13 and answered a real turn through the genuine Codex CLI. Served from the person's own ChatGPT plan through the Codex backend; no per-token price - the plan's 5-hour and weekly windows are what it costs. The Codex backend takes 272k of context. Codex ultra effort is refused by the backend.",
+          "defaultEffort": "medium"
         },
         {
           "id": "gpt-6-astra",
@@ -352,7 +363,8 @@
           "supportsThinking": true,
           "supportsVision": true,
           "effortLevels": ["low", "medium", "high", "xhigh", "max"],
-          "notes": "Listed by the Codex backend for the operator's ChatGPT account on 2026-10-13 and answered a real turn through the genuine Codex CLI. Served from the person's own ChatGPT plan through the Codex backend; no per-token price - the plan's 5-hour and weekly windows are what it costs. The Codex backend takes 272k of context. Codex ultra effort is refused by the backend."
+          "notes": "Listed by the Codex backend for the operator's ChatGPT account on 2026-10-13 and answered a real turn through the genuine Codex CLI. Served from the person's own ChatGPT plan through the Codex backend; no per-token price - the plan's 5-hour and weekly windows are what it costs. The Codex backend takes 272k of context. Codex ultra effort is refused by the backend.",
+          "defaultEffort": "low"
         },
         {
           "id": "gpt-5.6-sol",
@@ -362,7 +374,8 @@
           "supportsThinking": true,
           "supportsVision": true,
           "effortLevels": ["low", "medium", "high", "xhigh", "max"],
-          "notes": "Listed by the Codex backend for the operator's ChatGPT account on 2026-10-13 and answered a real turn through the genuine Codex CLI. Served from the person's own ChatGPT plan through the Codex backend; no per-token price - the plan's 5-hour and weekly windows are what it costs. The Codex backend takes 272k of context. Codex ultra effort is refused by the backend."
+          "notes": "Listed by the Codex backend for the operator's ChatGPT account on 2026-10-13 and answered a real turn through the genuine Codex CLI. Served from the person's own ChatGPT plan through the Codex backend; no per-token price - the plan's 5-hour and weekly windows are what it costs. The Codex backend takes 272k of context. Codex ultra effort is refused by the backend.",
+          "defaultEffort": "low"
         }
       ],
       "rateLimits": {
@@ -430,7 +443,8 @@
           "supportsVision": true,
           "effortLevels": ["low", "medium", "high", "xhigh", "max"],
           "maxOutputTokens": 128000,
-          "notes": "Codex's default (priority 4 in its model list). OpenAI API list price $0.10 / $0.01 cached / $0.50 per 1M (developers.openai.com/api/docs/models/gpt-6-luna, 2026-10-12) — the catalogue cost cells. The Codex backend takes 272k of its 1.05M window."
+          "notes": "Codex's default (priority 4 in its model list). OpenAI API list price $0.10 / $0.01 cached / $0.50 per 1M (developers.openai.com/api/docs/models/gpt-6-luna, 2026-10-12) — the catalogue cost cells. The Codex backend takes 272k of its 1.05M window.",
+          "defaultEffort": "medium"
         },
         {
           "id": "gpt-5.6-terra",
@@ -441,7 +455,8 @@
           "supportsVision": true,
           "effortLevels": ["low", "medium", "high", "xhigh", "max"],
           "maxOutputTokens": 128000,
-          "notes": "OpenAI API list price $2 / $0.20 cached / $12 per 1M (developers.openai.com/api/docs/models/gpt-5.6-terra, 2026-10-12). The Codex backend takes 272k."
+          "notes": "OpenAI API list price $2 / $0.20 cached / $12 per 1M (developers.openai.com/api/docs/models/gpt-5.6-terra, 2026-10-12). The Codex backend takes 272k.",
+          "defaultEffort": "medium"
         },
         {
           "id": "gpt-5.6-luna",
@@ -452,7 +467,8 @@
           "supportsVision": true,
           "effortLevels": ["low", "medium", "high", "xhigh", "max"],
           "maxOutputTokens": 128000,
-          "notes": "OpenAI API list price $0.20 / $0.02 cached / $1.20 per 1M (developers.openai.com/api/docs/models/gpt-5.6-luna, 2026-10-12). The Codex backend takes 272k."
+          "notes": "OpenAI API list price $0.20 / $0.02 cached / $1.20 per 1M (developers.openai.com/api/docs/models/gpt-5.6-luna, 2026-10-12). The Codex backend takes 272k.",
+          "defaultEffort": "medium"
         },
         {
           "id": "gpt-5.5",
@@ -463,7 +479,8 @@
           "supportsVision": true,
           "effortLevels": ["low", "medium", "high", "xhigh"],
           "maxOutputTokens": 128000,
-          "notes": "Legacy: Codex hides it from its picker, the backend still serves it. No max effort. OpenAI API list price $5 / $0.50 cached / $30 per 1M (developers.openai.com/api/docs/models/gpt-5.5, 2026-10-12)."
+          "notes": "Legacy: Codex hides it from its picker, the backend still serves it. No max effort. OpenAI API list price $5 / $0.50 cached / $30 per 1M (developers.openai.com/api/docs/models/gpt-5.5, 2026-10-12).",
+          "defaultEffort": "medium"
         },
         {
           "id": "gpt-6.1-sol",
@@ -474,7 +491,8 @@
           "supportsVision": true,
           "effortLevels": ["low", "medium", "high", "xhigh", "max"],
           "maxOutputTokens": 128000,
-          "notes": "Listed by the Codex backend for the operator's ChatGPT account on 2026-10-13 and answered a real turn through the genuine Codex CLI. OpenAI API list price $2 / $0.1 cached / $10 per 1M (developers.openai.com/api/docs/models/gpt-6.1-sol, 2026-10-13). The Codex backend takes 500k of context. Codex ultra effort is refused by the backend."
+          "notes": "Listed by the Codex backend for the operator's ChatGPT account on 2026-10-13 and answered a real turn through the genuine Codex CLI. OpenAI API list price $2 / $0.1 cached / $10 per 1M (developers.openai.com/api/docs/models/gpt-6.1-sol, 2026-10-13). The Codex backend takes 500k of context. Codex ultra effort is refused by the backend.",
+          "defaultEffort": "low"
         },
         {
           "id": "gpt-6-sol",
@@ -485,7 +503,8 @@
           "supportsVision": true,
           "effortLevels": ["low", "medium", "high", "xhigh", "max"],
           "maxOutputTokens": 128000,
-          "notes": "Listed by the Codex backend for the operator's ChatGPT account on 2026-10-13 and answered a real turn through the genuine Codex CLI. OpenAI API list price $2 / $0.2 cached / $10 per 1M (developers.openai.com/api/docs/models/gpt-6-sol, 2026-10-13). The Codex backend takes 272k of context. Codex ultra effort is refused by the backend."
+          "notes": "Listed by the Codex backend for the operator's ChatGPT account on 2026-10-13 and answered a real turn through the genuine Codex CLI. OpenAI API list price $2 / $0.2 cached / $10 per 1M (developers.openai.com/api/docs/models/gpt-6-sol, 2026-10-13). The Codex backend takes 272k of context. Codex ultra effort is refused by the backend.",
+          "defaultEffort": "medium"
         },
         {
           "id": "gpt-6-astra",
@@ -496,7 +515,8 @@
           "supportsVision": true,
           "effortLevels": ["low", "medium", "high", "xhigh", "max"],
           "maxOutputTokens": 128000,
-          "notes": "Listed by the Codex backend for the operator's ChatGPT account on 2026-10-13 and answered a real turn through the genuine Codex CLI. OpenAI API list price $10 / $1 cached / $50 per 1M (developers.openai.com/api/docs/models/gpt-6-astra, 2026-10-13). The Codex backend takes 272k of context. Codex ultra effort is refused by the backend."
+          "notes": "Listed by the Codex backend for the operator's ChatGPT account on 2026-10-13 and answered a real turn through the genuine Codex CLI. OpenAI API list price $10 / $1 cached / $50 per 1M (developers.openai.com/api/docs/models/gpt-6-astra, 2026-10-13). The Codex backend takes 272k of context. Codex ultra effort is refused by the backend.",
+          "defaultEffort": "low"
         },
         {
           "id": "gpt-5.6-sol",
@@ -507,7 +527,8 @@
           "supportsVision": true,
           "effortLevels": ["low", "medium", "high", "xhigh", "max"],
           "maxOutputTokens": 128000,
-          "notes": "Listed by the Codex backend for the operator's ChatGPT account on 2026-10-13 and answered a real turn through the genuine Codex CLI. OpenAI API list price $4 / $0.4 cached / $20 per 1M (developers.openai.com/api/docs/models/gpt-5.6-sol, 2026-10-13). The Codex backend takes 272k of context. Codex ultra effort is refused by the backend."
+          "notes": "Listed by the Codex backend for the operator's ChatGPT account on 2026-10-13 and answered a real turn through the genuine Codex CLI. OpenAI API list price $4 / $0.4 cached / $20 per 1M (developers.openai.com/api/docs/models/gpt-5.6-sol, 2026-10-13). The Codex backend takes 272k of context. Codex ultra effort is refused by the backend.",
+          "defaultEffort": "low"
         }
       ],
       "modelAliases": {
@@ -641,7 +662,7 @@
           "supportsTools": true,
           "supportsThinking": true,
           "supportsVision": true,
-          "effortLevels": ["low", "medium", "high", "xhigh", "max"],
+          "effortLevels": ["low", "medium", "high", "xhigh"],
           "maxOutputTokens": 128000,
           "notes": "Cheaper: Meta's contributor-mode price $0.10 input / $0.002 cached / $0.20 output per 1M tokens (dev.meta.ai/docs/pricing-rate-limits, 2026-10-13); contributor rate limit 150 RPM / 3M TPM. Contributor mode: Meta may use the content sent for product improvement (the owner accepted this for the pool, 2026-10-13); sent with the model id's -contributor suffix and the x-tbh-contributor-mode header."
         },
@@ -737,7 +758,8 @@
           },
           "notes": "Signed in by the server's own Muse CLI (version pinned in @oonacode/shared MUSE_TESTED_VERSION, installed at boot from Meta's download endpoint into the musedata volume, SHA-256-verified against Meta's manifest). The identity headers are the ones the real Muse sends (recorded 2026-10-13) — bump the user-agent version with Muse."
         }
-      ]
+      ],
+      "defaultEffort": "high"
     },
     {
       "id": "muse-subscription",
@@ -808,7 +830,7 @@
           "supportsTools": true,
           "supportsThinking": true,
           "supportsVision": true,
-          "effortLevels": ["low", "medium", "high", "xhigh", "max"],
+          "effortLevels": ["low", "medium", "high", "xhigh"],
           "maxOutputTokens": 128000,
           "notes": "Served from the person's own Muse subscription key through Meta's Responses API (api.meta.ai/v1/responses); the plan's 5-hour windows are what it costs. Cheaper: Meta's contributor-mode price $0.10 input / $0.002 cached / $0.20 output per 1M tokens (dev.meta.ai/docs/pricing-rate-limits, 2026-10-13); contributor rate limit 150 RPM / 3M TPM. Contributor mode: Meta may use the content sent for product improvement (the owner accepted this for the pool, 2026-10-13); sent with the model id's -contributor suffix and the x-tbh-contributor-mode header."
         },
@@ -873,7 +895,8 @@
       "adminGroup": "muse",
       "recordLabel": "Subscription",
       "docsUrl": "https://dev.meta.ai/docs/muse-code",
-      "notes": "The user's OWN Meta Muse key — a Muse Code subscription key (minted by `muse login`) or a Meta Model API key — kept in the vault and used straight from the PC (guardrail 4; the gateway is never on the path and this entry seeds no server rows). Muse speaks the OpenAI Responses API at api.meta.ai/v1/responses, so the engine reaches it through the host's loopback upstream, which translates every request (translate.wire = responses) and presents Muse's identity headers. Meta's terms: a subscription key is for use with Muse Code only — that is the person's standing, as for the other consumer plans. Models and limits are what Meta's own model list answered on 2026-10-13. docs/muse-engine.md."
+      "notes": "The user's OWN Meta Muse key — a Muse Code subscription key (minted by `muse login`) or a Meta Model API key — kept in the vault and used straight from the PC (guardrail 4; the gateway is never on the path and this entry seeds no server rows). Muse speaks the OpenAI Responses API at api.meta.ai/v1/responses, so the engine reaches it through the host's loopback upstream, which translates every request (translate.wire = responses) and presents Muse's identity headers. Meta's terms: a subscription key is for use with Muse Code only — that is the person's standing, as for the other consumer plans. Models and limits are what Meta's own model list answered on 2026-10-13. docs/muse-engine.md.",
+      "defaultEffort": "high"
     },
     {
       "id": "xai",
@@ -898,7 +921,8 @@
           "supportsVision": false,
           "inputPricePerM": 4,
           "outputPricePerM": 12,
-          "notes": "xAI's newest frontier model (SuperGrok's proxy lists it alone, 2026-10-04). Read from api.x.ai /v1/models with the pool's key that night: 500,000 context, prices identical to grok-4.6, reasoning effort low/medium/high/xhigh. TIERED upstream: this is the >=200k price; below 200k prompt tokens xAI charges half of it ($2 in / $0.50 cached / $6 out). The cheaper tier is `models.price_tier_rule` (migration 0163 copies grok-4.6's rows)."
+          "notes": "xAI's newest frontier model (SuperGrok's proxy lists it alone, 2026-10-04). Read from api.x.ai /v1/models with the pool's key that night: 500,000 context, prices identical to grok-4.6, reasoning effort low/medium/high/xhigh. TIERED upstream: this is the >=200k price; below 200k prompt tokens xAI charges half of it ($2 in / $0.50 cached / $6 out). The cheaper tier is `models.price_tier_rule` (migration 0163 copies grok-4.6's rows).",
+          "defaultEffort": "high"
         },
         {
           "id": "grok-4.6",
@@ -910,11 +934,12 @@
           "supportsVision": false,
           "inputPricePerM": 4,
           "outputPricePerM": 12,
-          "notes": "xAI's recommendation for code and chat. TIERED upstream: this is the >=200k price; below 200k prompt tokens xAI charges half of it. The cheaper tier is `models.price_tier_rule` (migration 0054)."
+          "notes": "xAI's recommendation for code and chat. TIERED upstream: this is the >=200k price; below 200k prompt tokens xAI charges half of it. The cheaper tier is `models.price_tier_rule` (migration 0054).",
+          "defaultEffort": "high"
         },
         {
           "id": "grok-4.5",
-          "effortLevels": ["low", "medium", "high", "xhigh"],
+          "effortLevels": ["low", "medium", "high"],
           "displayName": "Grok 4.5",
           "contextWindow": 500000,
           "supportsTools": true,
@@ -922,7 +947,8 @@
           "supportsVision": false,
           "inputPricePerM": 4,
           "outputPricePerM": 12,
-          "notes": "TIERED upstream: this is the >=200k price; below 200k prompt tokens xAI charges half of it. The cheaper tier is `models.price_tier_rule` (migration 0054)."
+          "notes": "TIERED upstream: this is the >=200k price; below 200k prompt tokens xAI charges half of it. The cheaper tier is `models.price_tier_rule` (migration 0054).",
+          "defaultEffort": "high"
         },
         {
           "id": "grok-4.3",
@@ -1067,7 +1093,8 @@
           "supportsThinking": true,
           "supportsVision": false,
           "inputPricePerM": 4,
-          "outputPricePerM": 12
+          "outputPricePerM": 12,
+          "defaultEffort": "high"
         },
         {
           "id": "grok-4.7-build-fast",
@@ -1078,7 +1105,8 @@
           "supportsThinking": true,
           "supportsVision": false,
           "inputPricePerM": 8,
-          "outputPricePerM": 24
+          "outputPricePerM": 24,
+          "defaultEffort": "high"
         },
         {
           "id": "grok-4.6",
@@ -1089,18 +1117,20 @@
           "supportsThinking": true,
           "supportsVision": false,
           "inputPricePerM": 4,
-          "outputPricePerM": 12
+          "outputPricePerM": 12,
+          "defaultEffort": "high"
         },
         {
           "id": "grok-4.5",
-          "effortLevels": ["low", "medium", "high", "xhigh"],
+          "effortLevels": ["low", "medium", "high"],
           "displayName": "Grok 4.5",
           "contextWindow": 500000,
           "supportsTools": true,
           "supportsThinking": true,
           "supportsVision": false,
           "inputPricePerM": 4,
-          "outputPricePerM": 12
+          "outputPricePerM": 12,
+          "defaultEffort": "high"
         },
         {
           "id": "grok-4.3",
@@ -1911,7 +1941,8 @@
           "notes": "The operator’s Antigravity account. The base URL is the daily Code Assist channel Antigravity itself talks to (its language server runs with --cloud_code_endpoint https://daily-cloudcode-pa.googleapis.com; prod is https://cloudcode-pa.googleapis.com — the row’s base URL decides). The ONE identity header is the user-agent Google’s own agy CLI sends (1.3.0, os_type=linux — the server’s own agy — cl=994000447, recorded 2026-10-06 on Windows and on the server’s Linux; docs/antigravity.md section 12) — bump its version with Antigravity’s, after the new agy has been recorded and tested on dev. Scopes: the six the language server carries minus `aicode`, which the public Antigravity sign-in integrations do not request; add it here if Google ever refuses a Code Assist call for want of it."
         }
       ],
-      "effortLevels": ["low", "medium", "high"]
+      "effortLevels": ["low", "medium", "high"],
+      "defaultEffort": "high"
     },
     {
       "id": "nvidia",