@oh-my-pi/pi-catalog 16.3.0 → 16.3.3

package/CHANGELOG.md

@@ -2,6 +2,26 @@
 
 ## [Unreleased]
 
+## [16.3.3] - 2026-07-02
+
+### Fixed
+
+- Extended Anthropic-compatible signing-endpoint recognition to Cloudflare AI Gateway, Google Vertex, AWS Bedrock, and Azure AI Inference / Foundry to ensure consistent reasoning-replay and signature-stripping behavior, and exposed ResolvedAnthropicCompat.signingEndpoint in the public API.
+- Fixed Zhipu Coding Plan runtime discovery to prioritize account-scoped model lists over bundled fallback models, preventing routing errors for valid non-z.ai keys.
+
+## [16.3.2] - 2026-07-02
+
+### Fixed
+
+- Fixed ZenMux model discovery to run without a `ZENMUX_API_KEY`, so newly published ZenMux models (for example `anthropic/claude-fable-5-free`) auto-update into the runtime `models.db` cache instead of waiting on a regenerated `models.json`.
+- Fixed ZenMux runtime discovery to query the `/api/v1/models` endpoint even when the resolved provider base URL points at the Anthropic-compatible route, so discovery no longer requests a non-existent `/api/anthropic/models` path.
+
+## [16.3.1] - 2026-07-02
+
+### Removed
+
+- Removed reasoning suppression prompt logic for GPT-5 models
+
 ## [16.3.0] - 2026-07-02
 
 ### Breaking Changes

package/dist/types/compat/openai.d.ts

@@ -20,8 +20,6 @@ interface OpenAIResponsesSpecLike {
  * as well as URL — bundled `azure` models carry no baseUrl (the deployment host
  * is per-resource, resolved at runtime) — while OpenAI/Copilot developer-role
  * and prompt-cache detection stay URL-keyed, as the historical call sites were.
- * The GPT-5 reasoning-suppression prompt keys on the model name, matching the
- * historical request-time check.
  */
 export declare function buildOpenAIResponsesCompat(spec: OpenAIResponsesSpecLike): ResolvedOpenAIResponsesCompat;
 export declare function buildOpenRouterCompat(spec: ModelSpec<"openrouter">): ResolvedOpenRouterCompat;

package/dist/types/provider-models/descriptors.d.ts

@@ -407,14 +407,17 @@ export declare const CATALOG_PROVIDERS: readonly [{
     readonly defaultModel: "anthropic/claude-opus-4.8";
     readonly envVars: readonly ["ZENMUX_API_KEY"];
     readonly createModelManagerOptions: (config: ModelManagerConfig) => import("..").ModelManagerOptions<import("..").Api, unknown>;
+    readonly allowUnauthenticated: true;
     readonly catalogDiscovery: {
         readonly label: "ZenMux";
+        readonly allowUnauthenticated: true;
     };
 }, {
     readonly id: "zhipu-coding-plan";
-    readonly defaultModel: "glm-5.2";
+    readonly defaultModel: "glm-5.1";
     readonly envVars: readonly ["ZHIPU_API_KEY"];
     readonly createModelManagerOptions: (config: ModelManagerConfig) => import("..").ModelManagerOptions<"openai-completions", unknown>;
+    readonly dynamicModelsAuthoritative: true;
     readonly catalogDiscovery: {
         readonly label: "Zhipu Coding Plan";
     };

package/dist/types/types.d.ts

@@ -286,14 +286,6 @@ export interface OpenAICompat {
     strictResponsesPairing?: boolean;
     /** Whether the Responses API accepts the `detail: "original"` image hint. Default: auto-detected (false for GitHub Copilot, which rejects it with a 400). */
     supportsImageDetailOriginal?: boolean;
-    /**
-     * Append a trailing no-reasoning developer item when the caller did not
-     * request reasoning, suppressing default reasoning on models that cannot
-     * disable it via request params (Responses APIs only; see
-     * https://community.openai.com/t/need-reasoning-false-option-for-gpt-5/1351588/7).
-     * The prompt must not look like an execution or tool budget. Default: auto-detected (GPT-5-family model names).
-     */
-    requiresReasoningSuppressionPrompt?: boolean;
     /** Whether streamed reasoning deltas for the same field may repeat the full cumulative text snapshot. Default: false. */
     reasoningDeltasMayBeCumulative?: boolean;
     /** Strip leaked DeepSeek chat-template special tokens from visible content deltas. Default: auto-detected. */
@@ -463,7 +455,7 @@ export interface ResolvedOpenAISharedCompat {
  * `buildModel`; request handlers read fields and never detect, resolve, or
  * allocate.
  */
-export type ResolvedOpenAICompat = ResolvedOpenAISharedCompat & Required<Omit<OpenAICompat, "supportsDeveloperRole" | "supportsReasoningEffort" | "reasoningEffortMap" | "supportsReasoningParams" | "thinkingFormat" | "reasoningDisableMode" | "omitReasoningEffort" | "includeEncryptedReasoning" | "filterReasoningHistory" | "disableReasoningOnForcedToolChoice" | "disableReasoningOnToolChoice" | "supportsToolChoice" | "supportsForcedToolChoice" | "supportsNamedToolChoice" | "reasoningContentField" | "requiresReasoningContentForToolCalls" | "requiresReasoningContentForAllAssistantTurns" | "allowsSyntheticReasoningContentForToolCalls" | "replayReasoningContent" | "qwenPreserveThinking" | "requiresThinkingAsText" | "requiresMistralToolIds" | "requiresToolResultName" | "requiresAssistantAfterToolResult" | "requiresAssistantContentForToolCalls" | "stripDeepseekSpecialTokens" | "streamMarkupHealingPattern" | "reasoningDeltasMayBeCumulative" | "emptyLengthFinishIsContextError" | "usesOpenAIToolCallIdLimit" | "promptCacheSessionHeader" | "openRouterRouting" | "isOpenRouterHost" | "supportsStrictMode" | "supportsLongPromptCacheRetention" | "alwaysSendMaxTokens" | "wireModelIdMode" | "vercelGatewayRouting" | "extraBody" | "toolStrictMode" | "toolSchemaFlavor" | "streamIdleTimeoutMs" | "cacheControlFormat" | "thinkingKeep" | "strictResponsesPairing" | "supportsImageDetailOriginal" | "requiresReasoningSuppressionPrompt" | "enableGeminiThinkingLoopGuard" | "whenThinking">> & {
+export type ResolvedOpenAICompat = ResolvedOpenAISharedCompat & Required<Omit<OpenAICompat, "supportsDeveloperRole" | "supportsReasoningEffort" | "reasoningEffortMap" | "supportsReasoningParams" | "thinkingFormat" | "reasoningDisableMode" | "omitReasoningEffort" | "includeEncryptedReasoning" | "filterReasoningHistory" | "disableReasoningOnForcedToolChoice" | "disableReasoningOnToolChoice" | "supportsToolChoice" | "supportsForcedToolChoice" | "supportsNamedToolChoice" | "reasoningContentField" | "requiresReasoningContentForToolCalls" | "requiresReasoningContentForAllAssistantTurns" | "allowsSyntheticReasoningContentForToolCalls" | "replayReasoningContent" | "qwenPreserveThinking" | "requiresThinkingAsText" | "requiresMistralToolIds" | "requiresToolResultName" | "requiresAssistantAfterToolResult" | "requiresAssistantContentForToolCalls" | "stripDeepseekSpecialTokens" | "streamMarkupHealingPattern" | "reasoningDeltasMayBeCumulative" | "emptyLengthFinishIsContextError" | "usesOpenAIToolCallIdLimit" | "promptCacheSessionHeader" | "openRouterRouting" | "isOpenRouterHost" | "supportsStrictMode" | "supportsLongPromptCacheRetention" | "alwaysSendMaxTokens" | "wireModelIdMode" | "vercelGatewayRouting" | "extraBody" | "toolStrictMode" | "toolSchemaFlavor" | "streamIdleTimeoutMs" | "cacheControlFormat" | "thinkingKeep" | "strictResponsesPairing" | "supportsImageDetailOriginal" | "enableGeminiThinkingLoopGuard" | "whenThinking">> & {
     vercelGatewayRouting?: OpenAICompat["vercelGatewayRouting"];
     extraBody?: OpenAICompat["extraBody"];
     cacheControlFormat?: OpenAICompat["cacheControlFormat"];
@@ -482,7 +474,6 @@ export interface ResolvedOpenAIResponsesCompat extends ResolvedOpenAISharedCompa
     supportsLongPromptCacheRetention: boolean;
     strictResponsesPairing: boolean;
     supportsImageDetailOriginal: boolean;
-    requiresReasoningSuppressionPrompt: boolean;
     supportsObfuscationOptOut: boolean;
     streamIdleTimeoutMs?: number;
 }
@@ -501,6 +492,17 @@ export type ResolvedAnthropicCompat = Required<AnthropicCompat> & {
      * env headers, and cache-TTL shaping without per-request URL parsing.
      */
     officialEndpoint: boolean;
+    /**
+     * The configured endpoint enforces Anthropic's signature protocol on
+     * replayed thinking blocks — either the official API itself or a proxy
+     * that forwards to it (GitHub Copilot, ZenMux, Cloudflare AI Gateway's
+     * `/anthropic` route, Google Vertex's `publishers/anthropic/…`).
+     * Downstream transforms strip stale cross-model thinking signatures on
+     * these endpoints so the signing proxy doesn't 400 with
+     * `Invalid signature in thinking block` (#4297). Superset of
+     * {@link officialEndpoint}.
+     */
+    signingEndpoint: boolean;
 };
 /**
  * Compatibility settings for the devin-agent (Codeium Cascade) API. Cascade

package/package.json

@@ -1,7 +1,7 @@
 {
 	"type": "module",
 	"name": "@oh-my-pi/pi-catalog",
-	"version": "16.3.0",
+	"version": "16.3.3",
 	"description": "Model catalog for omp: bundled model database, provider discovery descriptors, model identity, classification, and equivalence",
 	"homepage": "https://omp.sh",
 	"author": "Can Boluk",
@@ -34,12 +34,12 @@
 	},
 	"dependencies": {
 		"@bufbuild/protobuf": "^2.12.0",
-		"@oh-my-pi/pi-utils": "16.3.0",
+		"@oh-my-pi/pi-utils": "16.3.3",
 		"arktype": "^2.2.0",
 		"zod": "^4"
 	},
 	"devDependencies": {
-		"@oh-my-pi/pi-ai": "16.3.0",
+		"@oh-my-pi/pi-ai": "16.3.3",
 		"@types/bun": "^1.3.14"
 	},
 	"engines": {

package/src/compat/anthropic.ts

@@ -36,6 +36,52 @@ function matchesKimiK27CodeFamily(spec: ModelSpec<"anthropic-messages">): boolea
 	return spec.id === "kimi-for-coding" && /k2\.?7 code/i.test(spec.name ?? "");
 }
 
+const CLOUDFLARE_ANTHROPIC_GATEWAY_URL_MARKER = /gateway\.ai\.cloudflare\.com\/.+\/anthropic(?:\/|$)/i;
+const VERTEX_ANTHROPIC_URL_MARKER = /aiplatform\.googleapis\.com\/.+\/publishers\/anthropic\//i;
+const BEDROCK_ANTHROPIC_URL_MARKER = /(?:^|\/\/|\.)bedrock-runtime\.[a-z0-9-]+\.amazonaws\.com/i;
+const AZURE_ANTHROPIC_URL_MARKER = /(?:^|\/\/|\.)[a-z0-9-]+\.(?:inference|services)\.ai\.azure\.com/i;
+
+/**
+ * Cloudflare AI Gateway's `/anthropic` route forwards to signature-enforcing
+ * Anthropic (same failure class as GitHub Copilot #2851 / ZenMux #4192).
+ * Detection is by baseUrl marker rather than provider id: users routinely
+ * declare `provider: "custom"` (or other free-form ids) in `models.yml` for
+ * their own Cloudflare gateway account.
+ */
+function isCloudflareAnthropicGateway(baseUrl?: string): boolean {
+	return baseUrl !== undefined && CLOUDFLARE_ANTHROPIC_GATEWAY_URL_MARKER.test(baseUrl);
+}
+
+/**
+ * Google Vertex's `publishers/anthropic/models/…:streamRawPredict` route
+ * proxies Claude through Google's identity layer and returns full thinking
+ * signatures, so it is a SIGNING endpoint.
+ */
+function isVertexAnthropicRoute(baseUrl?: string): boolean {
+	return baseUrl !== undefined && VERTEX_ANTHROPIC_URL_MARKER.test(baseUrl);
+}
+
+/**
+ * AWS Bedrock's Anthropic route (`bedrock-runtime.<region>.amazonaws.com`)
+ * forwards Claude requests through Anthropic's signature protocol. Users can
+ * front Bedrock with a custom `anthropic-messages` provider entry in
+ * `models.yml`; the URL marker makes those signing by default without
+ * requiring a provider-id list.
+ */
+function isBedrockAnthropicRoute(baseUrl?: string): boolean {
+	return baseUrl !== undefined && BEDROCK_ANTHROPIC_URL_MARKER.test(baseUrl);
+}
+
+/**
+ * Azure AI Inference / Foundry Anthropic route
+ * (`<resource>.inference.ai.azure.com`, `<resource>.services.ai.azure.com`).
+ * Fronts Claude behind Azure identity and enforces Anthropic signatures on
+ * replay.
+ */
+function isAzureAnthropicRoute(baseUrl?: string): boolean {
+	return baseUrl !== undefined && AZURE_ANTHROPIC_URL_MARKER.test(baseUrl);
+}
+
 /** Build the resolved anthropic-messages compat record for a model spec. */
 export function buildAnthropicCompat(spec: ModelSpec<"anthropic-messages">): ResolvedAnthropicCompat {
 	const baseUrl = spec.baseUrl;
@@ -53,8 +99,17 @@ export function buildAnthropicCompat(spec: ModelSpec<"anthropic-messages">): Res
 	// (issue #4192).
 	const isZenmux = modelMatchesHost(spec, "zenmux");
 	const requiresThinkingEnabled = modelMatchesHost(spec, "moonshotNative") && matchesKimiK27CodeFamily(spec);
+	const signingEndpoint =
+		official ||
+		isCopilot ||
+		isZenmux ||
+		isCloudflareAnthropicGateway(baseUrl) ||
+		isVertexAnthropicRoute(baseUrl) ||
+		isBedrockAnthropicRoute(baseUrl) ||
+		isAzureAnthropicRoute(baseUrl);
 	const compat: ResolvedAnthropicCompat = {
 		officialEndpoint: official,
+		signingEndpoint,
 		disableStrictTools: false,
 		disableAdaptiveThinking: false,
 		supportsEagerToolInputStreaming: !isCopilot,
@@ -73,26 +128,23 @@ export function buildAnthropicCompat(spec: ModelSpec<"anthropic-messages">): Res
 		// into a class that reads `.id`.
 		requiresToolResultId: isZai,
 		requiresThinkingEnabled,
-		// Official Anthropic enforces signature-based thinking-chain integrity, so
-		// unsigned thinking blocks must stay text there. Anthropic-compatible
-		// reasoning endpoints commonly emit unsigned thinking blocks while still
-		// expecting them back as `type: "thinking"` on continuation; demoting them
-		// loses the reasoning chain and can destabilize the next tool-call
-		// arguments (#2005). Known non-signing hosts (Z.AI, DeepSeek) are also
-		// preserved for compatibility.
+		// Official Anthropic and Anthropic-compatible signing proxies enforce
+		// signature-based thinking-chain integrity, so unsigned thinking blocks
+		// must stay text there. Every other `anthropic-messages` reasoning
+		// endpoint replays unsigned thinking natively so the reasoning chain
+		// survives continuation and doesn't destabilize the next tool-call
+		// arguments (#2005, #2257, #2265, #3288, #3433, #3434). Opaque custom
+		// signing proxies remain the reporter's responsibility to mark with
+		// `compat.replayUnsignedThinking: false`; the transport surfaces a
+		// pointed remediation the first time the signing 400 fires (#4297).
 		//
-		// GitHub Copilot's `anthropic-messages` proxy and ZenMux's Anthropic route
-		// are excluded: both forward to signature-enforcing Anthropic and return
-		// full thinking signatures, so they are SIGNING endpoints. Replaying a
-		// stripped/unsigned thinking block as `signature: ""` there 400s the whole
-		// request ("Invalid signature") — most visibly when a checkpoint/branch-
-		// return turn's end_turn-bound signature is stripped on replay (issues
-		// #2851, #4192). Treating them like official Anthropic degrades such
-		// blocks to text instead, which the API accepts.
-		replayUnsignedThinking:
-			!isCopilot &&
-			!isZenmux &&
-			(isZai || modelMatchesHost(spec, "deepseekFamily") || (spec.reasoning && !official)),
+		// Known signing Anthropic-messages hosts (Copilot, ZenMux, Cloudflare
+		// AI Gateway `/anthropic`, Google Vertex `publishers/anthropic`, AWS
+		// Bedrock `bedrock-runtime.<region>.amazonaws.com`, and Azure
+		// AI Inference / Foundry `<res>.(inference|services).ai.azure.com`)
+		// are excluded automatically because they can be recognised by provider
+		// id or baseUrl marker.
+		replayUnsignedThinking: !signingEndpoint && (Boolean(spec.reasoning) || modelMatchesHost(spec, "deepseekFamily")),
 		escapeBuiltinToolNames: modelMatchesHost(spec, "umans"),
 	};
 	applyCompatOverrides(compat, spec.compat);

package/src/compat/openai.ts

@@ -593,8 +593,6 @@ interface OpenAIResponsesSpecLike {
  * as well as URL — bundled `azure` models carry no baseUrl (the deployment host
  * is per-resource, resolved at runtime) — while OpenAI/Copilot developer-role
  * and prompt-cache detection stay URL-keyed, as the historical call sites were.
- * The GPT-5 reasoning-suppression prompt keys on the model name, matching the
- * historical request-time check.
  */
 export function buildOpenAIResponsesCompat(spec: OpenAIResponsesSpecLike): ResolvedOpenAIResponsesCompat {
 	const baseUrl = spec.baseUrl ?? "";
@@ -626,7 +624,6 @@ export function buildOpenAIResponsesCompat(spec: OpenAIResponsesSpecLike): Resol
 		// or base-URL host (mirroring the Anthropic compat builder) so a model
 		// pointed at the Copilot host under a different provider id still clamps.
 		supportsImageDetailOriginal: !modelMatchesHost({ provider: spec.provider, baseUrl }, "githubCopilot"),
-		requiresReasoningSuppressionPrompt: spec.name.toLowerCase().startsWith("gpt-5"),
 		reasoningEffortMap: {},
 		supportsReasoningParams: true,
 		thinkingFormat,
@@ -693,7 +690,6 @@ function pickResponsesOnly(compat: ResolvedOpenAIResponsesCompat): ResponsesOnly
 		supportsLongPromptCacheRetention: compat.supportsLongPromptCacheRetention,
 		strictResponsesPairing: compat.strictResponsesPairing,
 		supportsImageDetailOriginal: compat.supportsImageDetailOriginal,
-		requiresReasoningSuppressionPrompt: compat.requiresReasoningSuppressionPrompt,
 		supportsObfuscationOptOut: compat.supportsObfuscationOptOut,
 	} satisfies ResponsesOnlyCompat;
 }

package/src/models.json

@@ -8330,10 +8330,10 @@
 				"image"
 			],
 			"cost": {
-				"input": 11,
-				"output": 55,
-				"cacheRead": 1.1,
-				"cacheWrite": 13.75
+				"input": 10,
+				"output": 50,
+				"cacheRead": 1,
+				"cacheWrite": 12.5
 			},
 			"contextWindow": 1000000,
 			"maxTokens": 128000,
@@ -12658,23 +12658,33 @@
 	"cerebras": {
 		"gemma-4-31b": {
 			"id": "gemma-4-31b",
-			"name": "gemma-4-31b",
+			"name": "Gemma 4 31B IT",
 			"api": "openai-completions",
 			"provider": "cerebras",
 			"baseUrl": "https://api.cerebras.ai/v1",
-			"reasoning": false,
+			"reasoning": true,
 			"input": [
 				"text",
 				"image"
 			],
 			"cost": {
-				"input": 0,
-				"output": 0,
+				"input": 0.99,
+				"output": 1.49,
 				"cacheRead": 0,
 				"cacheWrite": 0
 			},
-			"contextWindow": 256000,
-			"maxTokens": 8192
+			"contextWindow": 131072,
+			"maxTokens": 40960,
+			"thinking": {
+				"mode": "effort",
+				"efforts": [
+					"minimal",
+					"low",
+					"medium",
+					"high",
+					"xhigh"
+				]
+			}
 		},
 		"gpt-oss-120b": {
 			"id": "gpt-oss-120b",
@@ -17558,7 +17568,7 @@
 		},
 		"minimax-m3": {
 			"id": "minimax-m3",
-			"name": "MiniMax-M3 (3x usage)",
+			"name": "MiniMax-M3",
 			"api": "openai-completions",
 			"provider": "fireworks",
 			"baseUrl": "https://api.fireworks.ai/inference/v1",
@@ -18096,6 +18106,48 @@
 				}
 			}
 		},
+		"claude-sonnet-5": {
+			"id": "claude-sonnet-5",
+			"name": "Claude Sonnet 5",
+			"api": "anthropic-messages",
+			"provider": "github-copilot",
+			"baseUrl": "https://api.githubcopilot.com",
+			"reasoning": true,
+			"input": [
+				"text",
+				"image"
+			],
+			"cost": {
+				"input": 2,
+				"output": 10,
+				"cacheRead": 0.2,
+				"cacheWrite": 2.5
+			},
+			"contextWindow": 1000000,
+			"maxTokens": 128000,
+			"headers": {
+				"User-Agent": "opencode/1.3.15",
+				"X-GitHub-Api-Version": "2026-06-01"
+			},
+			"thinking": {
+				"mode": "anthropic-adaptive",
+				"efforts": [
+					"minimal",
+					"low",
+					"medium",
+					"high",
+					"xhigh"
+				],
+				"effortMap": {
+					"minimal": "low",
+					"low": "medium",
+					"medium": "high",
+					"high": "xhigh",
+					"xhigh": "max"
+				},
+				"supportsDisplay": true
+			}
+		},
 		"gemini-2.5-pro": {
 			"id": "gemini-2.5-pro",
 			"name": "Gemini 2.5 Pro",
@@ -20148,7 +20200,7 @@
 		},
 		"gemma-4-31b": {
 			"id": "gemma-4-31b",
-			"name": "Gemma 4 31B",
+			"name": "Gemma 4 31B IT",
 			"api": "google-generative-ai",
 			"provider": "google",
 			"baseUrl": "https://generativelanguage.googleapis.com/v1beta",
@@ -60067,7 +60119,7 @@
 		},
 		"minimax-m3": {
 			"id": "minimax-m3",
-			"name": "MiniMax-M3 (3x usage)",
+			"name": "MiniMax-M3",
 			"api": "openai-completions",
 			"provider": "opencode-go",
 			"baseUrl": "https://opencode.ai/zen/go/v1",
@@ -63775,8 +63827,8 @@
 				"text"
 			],
 			"cost": {
-				"input": 0.19999999999999998,
-				"output": 0.77,
+				"input": 0.24,
+				"output": 0.8999999999999999,
 				"cacheRead": 0.135,
 				"cacheWrite": 0
 			},
@@ -64032,9 +64084,9 @@
 				"text"
 			],
 			"cost": {
-				"input": 0.098,
-				"output": 0.196,
-				"cacheRead": 0.02,
+				"input": 0.08900000000000001,
+				"output": 0.18,
+				"cacheRead": 0.018,
 				"cacheWrite": 0
 			},
 			"contextWindow": 1048576,
@@ -65283,12 +65335,12 @@
 			],
 			"cost": {
 				"input": 0.255,
-				"output": 1,
+				"output": 1.02,
 				"cacheRead": 0.03,
 				"cacheWrite": 0
 			},
 			"contextWindow": 204800,
-			"maxTokens": 196608,
+			"maxTokens": 131072,
 			"thinking": {
 				"mode": "effort",
 				"efforts": [
@@ -65310,13 +65362,13 @@
 				"text"
 			],
 			"cost": {
-				"input": 0.29,
-				"output": 0.95,
+				"input": 0.3,
+				"output": 1.2,
 				"cacheRead": 0.03,
 				"cacheWrite": 0
 			},
 			"contextWindow": 204800,
-			"maxTokens": 196608,
+			"maxTokens": 131072,
 			"thinking": {
 				"mode": "effort",
 				"efforts": [
@@ -66037,11 +66089,11 @@
 			"cost": {
 				"input": 0.6,
 				"output": 2.5,
-				"cacheRead": 0.6,
+				"cacheRead": 0.15,
 				"cacheWrite": 0
 			},
 			"contextWindow": 262144,
-			"maxTokens": 262144,
+			"maxTokens": 100352,
 			"thinking": {
 				"mode": "effort",
 				"efforts": [
@@ -68788,13 +68840,13 @@
 				"text"
 			],
 			"cost": {
-				"input": 0.08,
-				"output": 0.39999999999999997,
+				"input": 0.13,
+				"output": 1.56,
 				"cacheRead": 0.08,
 				"cacheWrite": 0
 			},
 			"contextWindow": 131072,
-			"maxTokens": 131072,
+			"maxTokens": 32768,
 			"thinking": {
 				"mode": "effort",
 				"efforts": [
@@ -68901,8 +68953,8 @@
 				"text"
 			],
 			"cost": {
-				"input": 0.049999999999999996,
-				"output": 0.39999999999999997,
+				"input": 0.117,
+				"output": 0.45499999999999996,
 				"cacheRead": 0.049999999999999996,
 				"cacheWrite": 0
 			},
@@ -71329,7 +71381,7 @@
 	"synthetic": {
 		"hf:MiniMaxAI/MiniMax-M3": {
 			"id": "hf:MiniMaxAI/MiniMax-M3",
-			"name": "MiniMaxAI/MiniMax-M3",
+			"name": "MiniMax-M3",
 			"api": "openai-completions",
 			"provider": "synthetic",
 			"baseUrl": "https://api.synthetic.new/openai/v1",
@@ -71344,7 +71396,7 @@
 				"cacheRead": 0.6,
 				"cacheWrite": 0
 			},
-			"contextWindow": 262144,
+			"contextWindow": 524288,
 			"maxTokens": 65536,
 			"thinking": {
 				"mode": "effort",
@@ -71389,26 +71441,37 @@
 		},
 		"hf:moonshotai/Kimi-K2.7-Code": {
 			"id": "hf:moonshotai/Kimi-K2.7-Code",
-			"name": "moonshotai/Kimi-K2.7-Code",
+			"name": "Kimi K2.7 Code",
 			"api": "openai-completions",
 			"provider": "synthetic",
 			"baseUrl": "https://api.synthetic.new/openai/v1",
-			"reasoning": false,
+			"reasoning": true,
 			"input": [
-				"text"
+				"text",
+				"image"
 			],
 			"cost": {
-				"input": 0,
-				"output": 0,
-				"cacheRead": 0,
+				"input": 0.95,
+				"output": 4,
+				"cacheRead": 0.95,
 				"cacheWrite": 0
 			},
 			"contextWindow": 262144,
-			"maxTokens": 8192
+			"maxTokens": 65536,
+			"thinking": {
+				"mode": "effort",
+				"efforts": [
+					"minimal",
+					"low",
+					"medium",
+					"high",
+					"xhigh"
+				]
+			}
 		},
 		"hf:nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-NVFP4": {
 			"id": "hf:nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-NVFP4",
-			"name": "nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-NVFP4",
+			"name": "Nemotron 3 Super 120B A12B",
 			"api": "openai-completions",
 			"provider": "synthetic",
 			"baseUrl": "https://api.synthetic.new/openai/v1",
@@ -71437,7 +71500,7 @@
 		},
 		"hf:openai/gpt-oss-120b": {
 			"id": "hf:openai/gpt-oss-120b",
-			"name": "openai/gpt-oss-120b",
+			"name": "GPT OSS 120B",
 			"api": "openai-completions",
 			"provider": "synthetic",
 			"baseUrl": "https://api.synthetic.new/openai/v1",
@@ -71464,7 +71527,7 @@
 		},
 		"hf:Qwen/Qwen3.6-27B": {
 			"id": "hf:Qwen/Qwen3.6-27B",
-			"name": "Qwen/Qwen3.6-27B",
+			"name": "Qwen3.6 27B",
 			"api": "openai-completions",
 			"provider": "synthetic",
 			"baseUrl": "https://api.synthetic.new/openai/v1",
@@ -71493,7 +71556,7 @@
 		},
 		"hf:zai-org/GLM-4.7-Flash": {
 			"id": "hf:zai-org/GLM-4.7-Flash",
-			"name": "zai-org/GLM-4.7-Flash",
+			"name": "GLM-4.7-Flash",
 			"api": "openai-completions",
 			"provider": "synthetic",
 			"baseUrl": "https://api.synthetic.new/openai/v1",
@@ -71551,7 +71614,7 @@
 		},
 		"hf:zai-org/GLM-5.2": {
 			"id": "hf:zai-org/GLM-5.2",
-			"name": "zai-org/GLM-5.2",
+			"name": "GLM-5.2",
 			"api": "openai-completions",
 			"provider": "synthetic",
 			"baseUrl": "https://api.synthetic.new/openai/v1",
@@ -73126,6 +73189,36 @@
 				"supportsUsageInStreaming": false
 			}
 		},
+		"claude-sonnet-5": {
+			"id": "claude-sonnet-5",
+			"name": "Claude Sonnet 5",
+			"api": "openai-completions",
+			"provider": "venice",
+			"baseUrl": "https://api.venice.ai/api/v1",
+			"reasoning": true,
+			"input": [
+				"text",
+				"image"
+			],
+			"cost": {
+				"input": 3,
+				"output": 15,
+				"cacheRead": 0.3,
+				"cacheWrite": 3.75
+			},
+			"contextWindow": 1000000,
+			"maxTokens": 64000,
+			"thinking": {
+				"mode": "effort",
+				"efforts": [
+					"minimal",
+					"low",
+					"medium",
+					"high",
+					"xhigh"
+				]
+			}
+		},
 		"deepseek-v3.2": {
 			"id": "deepseek-v3.2",
 			"name": "DeepSeek V3.2",
@@ -74395,7 +74488,7 @@
 		},
 		"minimax-m3": {
 			"id": "minimax-m3",
-			"name": "MiniMax-M3 (3x usage)",
+			"name": "MiniMax-M3",
 			"api": "openai-completions",
 			"provider": "venice",
 			"baseUrl": "https://api.venice.ai/api/v1",
@@ -82295,6 +82388,14 @@
 			},
 			"contextWindow": 2000000,
 			"maxTokens": 2000000,
+			"compat": {
+				"reasoningEffortMap": {
+					"minimal": "low"
+				},
+				"includeEncryptedReasoning": false,
+				"filterReasoningHistory": true,
+				"omitReasoningEffort": false
+			},
 			"thinking": {
 				"mode": "effort",
 				"efforts": [
@@ -82307,14 +82408,6 @@
 				"effortMap": {
 					"minimal": "low"
 				}
-			},
-			"compat": {
-				"reasoningEffortMap": {
-					"minimal": "low"
-				},
-				"includeEncryptedReasoning": false,
-				"filterReasoningHistory": true,
-				"omitReasoningEffort": false
 			}
 		},
 		"grok-4.3": {
@@ -82336,6 +82429,14 @@
 			},
 			"contextWindow": 1000000,
 			"maxTokens": 1000000,
+			"compat": {
+				"reasoningEffortMap": {
+					"minimal": "low"
+				},
+				"includeEncryptedReasoning": false,
+				"filterReasoningHistory": true,
+				"omitReasoningEffort": false
+			},
 			"thinking": {
 				"mode": "effort",
 				"efforts": [
@@ -82348,14 +82449,6 @@
 				"effortMap": {
 					"minimal": "low"
 				}
-			},
-			"compat": {
-				"reasoningEffortMap": {
-					"minimal": "low"
-				},
-				"includeEncryptedReasoning": false,
-				"filterReasoningHistory": true,
-				"omitReasoningEffort": false
 			}
 		},
 		"grok-build": {

package/src/provider-models/descriptors.ts

@@ -453,13 +453,15 @@ export const CATALOG_PROVIDERS = [
 		defaultModel: "anthropic/claude-opus-4.8",
 		envVars: ["ZENMUX_API_KEY"],
 		createModelManagerOptions: (config: ModelManagerConfig) => zenmuxModelManagerOptions(config),
-		catalogDiscovery: { label: "ZenMux" },
+		allowUnauthenticated: true,
+		catalogDiscovery: { label: "ZenMux", allowUnauthenticated: true },
 	},
 	{
 		id: "zhipu-coding-plan",
-		defaultModel: "glm-5.2",
+		defaultModel: "glm-5.1",
 		envVars: ["ZHIPU_API_KEY"],
 		createModelManagerOptions: (config: ModelManagerConfig) => zhipuCodingPlanModelManagerOptions(config),
+		dynamicModelsAuthoritative: true,
 		catalogDiscovery: { label: "Zhipu Coding Plan" },
 	},
 ] as const satisfies readonly ProviderCatalogEntry[];

package/src/provider-models/openai-compat.ts

@@ -1283,6 +1283,7 @@ export function zhipuCodingPlanModelManagerOptions(
 	const baseUrl = config?.baseUrl ?? "https://open.bigmodel.cn/api/coding/paas/v4";
 	return {
 		providerId: "zhipu-coding-plan",
+		dynamicModelsAuthoritative: true,
 		...(apiKey && {
 			fetchDynamicModels: () =>
 				fetchOpenAICompatibleModels({
@@ -2035,7 +2036,11 @@ function normalizeZenMuxOpenAiBaseUrl(baseUrl?: string): string {
 	if (!value) {
 		return ZENMUX_OPENAI_BASE_URL;
 	}
-	return value.endsWith("/") ? value.slice(0, -1) : value;
+	const normalized = value.endsWith("/") ? value.slice(0, -1) : value;
+	if (normalized.endsWith("/api/anthropic")) {
+		return normalized.replace("/api/anthropic", "/api/v1");
+	}
+	return normalized;
 }
 
 function toZenMuxAnthropicBaseUrl(openAiBaseUrl: string): string {
@@ -2103,37 +2108,35 @@ export function zenmuxModelManagerOptions(config?: ZenMuxModelManagerConfig): Mo
 	const anthropicBaseUrl = toZenMuxAnthropicBaseUrl(openAiBaseUrl);
 	return {
 		providerId: "zenmux",
-		...(apiKey && {
-			fetchDynamicModels: () =>
-				fetchOpenAICompatibleModels<Api>({
-					api: "openai-completions",
-					provider: "zenmux",
-					baseUrl: openAiBaseUrl,
-					apiKey,
-					mapModel: (entry, defaults) => {
-						const pricings = isRecord(entry.pricings) ? entry.pricings : undefined;
-						const capabilities = isRecord(entry.capabilities) ? entry.capabilities : undefined;
-						const isAnthropicModel = isZenMuxAnthropicModel(entry, defaults.id);
-						return {
-							...defaults,
-							name: toModelName(entry.display_name, defaults.name),
-							api: isAnthropicModel ? "anthropic-messages" : "openai-completions",
-							baseUrl: isAnthropicModel ? anthropicBaseUrl : openAiBaseUrl,
-							reasoning: capabilities?.reasoning === true || defaults.reasoning,
-							input: toInputCapabilities(entry.input_modalities),
-							cost: {
-								input: getZenMuxPricingValue(pricings, "prompt"),
-								output: getZenMuxPricingValue(pricings, "completion"),
-								cacheRead: getZenMuxPricingValue(pricings, "input_cache_read"),
-								cacheWrite: getZenMuxCacheWritePrice(pricings),
-							},
-							contextWindow: toPositiveNumber(entry.context_length, defaults.contextWindow),
-							maxTokens: toPositiveNumber(entry.max_completion_tokens, defaults.maxTokens),
-						};
-					},
-					fetch: config?.fetch,
-				}),
-		}),
+		fetchDynamicModels: () =>
+			fetchOpenAICompatibleModels<Api>({
+				api: "openai-completions",
+				provider: "zenmux",
+				baseUrl: openAiBaseUrl,
+				apiKey,
+				mapModel: (entry, defaults) => {
+					const pricings = isRecord(entry.pricings) ? entry.pricings : undefined;
+					const capabilities = isRecord(entry.capabilities) ? entry.capabilities : undefined;
+					const isAnthropicModel = isZenMuxAnthropicModel(entry, defaults.id);
+					return {
+						...defaults,
+						name: toModelName(entry.display_name, defaults.name),
+						api: isAnthropicModel ? "anthropic-messages" : "openai-completions",
+						baseUrl: isAnthropicModel ? anthropicBaseUrl : openAiBaseUrl,
+						reasoning: capabilities?.reasoning === true || defaults.reasoning,
+						input: toInputCapabilities(entry.input_modalities),
+						cost: {
+							input: getZenMuxPricingValue(pricings, "prompt"),
+							output: getZenMuxPricingValue(pricings, "completion"),
+							cacheRead: getZenMuxPricingValue(pricings, "input_cache_read"),
+							cacheWrite: getZenMuxCacheWritePrice(pricings),
+						},
+						contextWindow: toPositiveNumber(entry.context_length, defaults.contextWindow),
+						maxTokens: toPositiveNumber(entry.max_completion_tokens, defaults.maxTokens),
+					};
+				},
+				fetch: config?.fetch,
+			}),
 	};
 }
 

package/src/types.ts

@@ -324,14 +324,6 @@ export interface OpenAICompat {
 	strictResponsesPairing?: boolean;
 	/** Whether the Responses API accepts the `detail: "original"` image hint. Default: auto-detected (false for GitHub Copilot, which rejects it with a 400). */
 	supportsImageDetailOriginal?: boolean;
-	/**
-	 * Append a trailing no-reasoning developer item when the caller did not
-	 * request reasoning, suppressing default reasoning on models that cannot
-	 * disable it via request params (Responses APIs only; see
-	 * https://community.openai.com/t/need-reasoning-false-option-for-gpt-5/1351588/7).
-	 * The prompt must not look like an execution or tool budget. Default: auto-detected (GPT-5-family model names).
-	 */
-	requiresReasoningSuppressionPrompt?: boolean;
 	/** Whether streamed reasoning deltas for the same field may repeat the full cumulative text snapshot. Default: false. */
 	reasoningDeltasMayBeCumulative?: boolean;
 	/** Strip leaked DeepSeek chat-template special tokens from visible content deltas. Default: auto-detected. */
@@ -557,7 +549,6 @@ export type ResolvedOpenAICompat = ResolvedOpenAISharedCompat &
 			| "thinkingKeep"
 			| "strictResponsesPairing"
 			| "supportsImageDetailOriginal"
-			| "requiresReasoningSuppressionPrompt"
 			| "enableGeminiThinkingLoopGuard"
 			| "whenThinking"
 		>
@@ -581,7 +572,6 @@ export interface ResolvedOpenAIResponsesCompat extends ResolvedOpenAISharedCompa
 	supportsLongPromptCacheRetention: boolean;
 	strictResponsesPairing: boolean;
 	supportsImageDetailOriginal: boolean;
-	requiresReasoningSuppressionPrompt: boolean;
 	supportsObfuscationOptOut: boolean;
 	streamIdleTimeoutMs?: number;
 }
@@ -602,6 +592,17 @@ export type ResolvedAnthropicCompat = Required<AnthropicCompat> & {
 	 * env headers, and cache-TTL shaping without per-request URL parsing.
 	 */
 	officialEndpoint: boolean;
+	/**
+	 * The configured endpoint enforces Anthropic's signature protocol on
+	 * replayed thinking blocks — either the official API itself or a proxy
+	 * that forwards to it (GitHub Copilot, ZenMux, Cloudflare AI Gateway's
+	 * `/anthropic` route, Google Vertex's `publishers/anthropic/…`).
+	 * Downstream transforms strip stale cross-model thinking signatures on
+	 * these endpoints so the signing proxy doesn't 400 with
+	 * `Invalid signature in thinking block` (#4297). Superset of
+	 * {@link officialEndpoint}.
+	 */
+	signingEndpoint: boolean;
 };
 
 /**