@oh-my-pi/pi-catalog 15.12.4 → 15.13.1

package/CHANGELOG.md

@@ -2,6 +2,36 @@
 
 ## [Unreleased]
 
+## [15.13.1] - 2026-06-15
+
+### Added
+
+- Added `modelFamilyToken(modelId)` to `@oh-my-pi/pi-catalog/identity`: a coarse vendor-lineage token (`anthropic`/`openai`/`gemini`/`kimi`/…) for "are two models the same family?" comparisons, backed by `parseKnownModel` canonical-id normalization. Opaque and comparison-only; kind/variant collapsed onto the vendor token ([#2406](https://github.com/can1357/oh-my-pi/issues/2406))
+
+### Changed
+
+- Changed catalog metadata to update a model’s per-token pricing to input 0.09 and output 0.18
+- Changed the same cataloged model’s maximum token limit from 384000 to 65536
+
+### Fixed
+
+- Fixed MiniMax-M3 catalog context for `minimax` and `minimax-cn` to report the documented 1M long-context tier instead of the upstream 512K pricing boundary ([#2576](https://github.com/can1357/oh-my-pi/issues/2576)).
+- Fixed OpenCode Go MiMo catalog metadata so title generation and other tool-enabled calls omit unsupported `tool_choice` instead of triggering provider 400s ([#2509](https://github.com/can1357/oh-my-pi/issues/2509)).
+- Fixed OpenCode Go `kimi-k2.7-code` catalog metadata so resolve-gate requests use automatic tool selection instead of Moonshot-rejected forced `tool_choice` ([#2546](https://github.com/can1357/oh-my-pi/issues/2546)).
+- Fixed Anthropic compat for the `github-copilot` host so `supportsEagerToolInputStreaming` defaults to `false` there, matching the Copilot proxy which rejects the per-tool `eager_input_streaming` field ([#2558](https://github.com/can1357/oh-my-pi/issues/2558)).
+- Scoped vLLM model cache validity to the discovery base URL so changed endpoints refetch immediately, and bounded built-in vLLM discovery requests with a timeout.
+
+## [15.12.6] - 2026-06-14
+
+### Added
+
+- Added GLM-5.2 to the bundled zai (GLM Coding Plan) catalog as the selectable 1M served model.
+
+### Changed
+
+- Pinned zai `glm-5.2` to 1M context during catalog generation so endpoint discovery and older fallbacks cannot regress it to 200k.
+- Replaced the hand-maintained `zhipu-coding-plan` GLM reasoning allowlist and vision regex with a `parseGlmModel` family classifier in `identity/classify.ts` (variant + vision + version), surfaced as `isReasoningGlmModelId` / `isGlmVisionModelId`. Discovery now derives reasoning/vision capability from the GLM family instead of a per-id list, so newly-bumped integers (`glm-5.3`, `glm-6`, …) are covered automatically while `-flash`/`-preview` and the vision `…v` shape stay correctly classified.
+
 ## [15.12.4] - 2026-06-13
 
 ### Added
@@ -28,6 +58,7 @@
 - Fixed Antigravity `gemini-3.1-pro --thinking high` failing with `Cloud Code Assist API error (400): Request contains an invalid argument.` — the upstream `gemini-3.1-pro-high` deployment rejects every `streamGenerateContent` request on both CCA endpoints while discovery still advertises it. High effort now routes to `gemini-pro-agent` (the same "Gemini 3.1 Pro (High)" model, verified accepting the identical request body), and the model-cache fingerprint version was bumped (`merge-v2` → `merge-v3`) so existing fresh caches refetch discovery and pick up the corrected routing immediately.
 
 ## [15.11.7] - 2026-06-12
+
 ### Added
 
 - Added effort-tier variant collapsing (`variant-collapse`): providers that expose one logical model as several effort/thinking-suffixed upstream ids (Antigravity CCA `gemini-3.5-flash-extra-low`/`-low`/`gemini-3-flash-agent`, `gemini-3[.1]-pro-low|high`, `claude-*[-thinking]` pairs, `gpt-oss-120b-medium`) collapse into one logical entry carrying per-effort upstream routing in `thinking.effortRouting` (plus `thinking.suppressWhenOff` for Cloud Code Assist ids whose baked server default re-applies when `thinkingConfig` is omitted). Request-time code resolves the outbound id via `resolveWireModelId(model, effort)`; selection, caching, and usage attribution key on the logical id.
@@ -52,11 +83,13 @@
 ### Fixed
 
 - Fixed MiniMax M2-family and OpenAI gpt-oss model metadata so OpenAI-compatible catalog entries declare only `low|medium|high` thinking efforts. Their upstreams reject `minimal`, `xhigh`, and Fireworks' `minimal → none` wire mapping, so `fireworks/minimax-m2.7` as the smol auto-thinking classifier model 400ed on every turn. OpenAI-compatible provider effort maps (`Groq qwen/qwen3-32b`, DeepSeek-family, OpenRouter Anthropic adaptive, Fireworks `minimal → none`) now bake into `thinking.effortMap` in catalog metadata instead of `buildOpenAICompat`, and request builders read that field directly. Regenerated `models.json` now makes `disableReasoning` choose `low` for those families while leaving GLM-5.x and other Fireworks models on the existing `minimal → none` path ([#2315](https://github.com/can1357/oh-my-pi/issues/2315)).
+
 ### Added
 
 - Added `requiresJuiceZeroHack` Responses-API compat flag, resolved by `buildOpenAIResponsesCompat` from GPT-5-family model names and overridable via sparse model `compat` config. Replaces the request-time `model.name.startsWith("gpt-5")` sniff that gated the trailing `# Juice: 0 !important` no-reasoning developer item.
 
 ## [15.11.3] - 2026-06-11
+
 ### Added
 
 - Added `requestModelId` on `Model` to represent the upstream model id used when a catalog entry is a local variant
@@ -134,4 +167,4 @@
 
 ### Removed
 
-- Removed the runtime enrichment layer: `enrichModelThinking` (and its non-enumerable memo-slot cache), `refreshModelThinking`, `modelOmitsReasoningEffort`, and the `model-thinking` re-exports of generator-only policies. Thinking metadata is resolved exactly once inside `buildModel`; runtime helpers (`getSupportedEfforts`, `clampThinkingLevelForModel`, `requireSupportedEffort`, the effort mappers) are pure field reads.
+- Removed the runtime enrichment layer: `enrichModelThinking` (and its non-enumerable memo-slot cache), `refreshModelThinking`, `modelOmitsReasoningEffort`, and the `model-thinking` re-exports of generator-only policies. Thinking metadata is resolved exactly once inside `buildModel`; runtime helpers (`getSupportedEfforts`, `clampThinkingLevelForModel`, `requireSupportedEffort`, the effort mappers) are pure field reads.

package/dist/types/identity/classify.d.ts

@@ -12,6 +12,7 @@ export type SemVer = {
 export type GeminiKind = "pro" | "flash";
 export type AnthropicKind = "opus" | "sonnet" | "fable" | "mythos";
 export type OpenAIVariant = "base" | "codex" | "codex-max" | "codex-mini" | "codex-spark" | "mini" | "max" | "nano";
+export type GlmVariant = "base" | "air" | "turbo" | "flash" | "flashx" | "preview";
 export interface GeminiModel {
     family: "gemini";
     kind: GeminiKind;
@@ -27,6 +28,14 @@ export interface OpenAIModel {
     variant: OpenAIVariant;
     version: SemVer;
 }
+export interface GlmModel {
+    family: "glm";
+    /** Suffix variant (`-air`, `-turbo`, `-flash`, `-flashx`, `-preview`); `base` when none. */
+    variant: GlmVariant;
+    /** Vision SKU — the `v` that attaches directly to the version (`glm-4v`, `glm-4.5v`). */
+    vision: boolean;
+    version: SemVer;
+}
 export interface UnknownModel {
     family: "unknown";
     id: string;
@@ -35,9 +44,18 @@ export type ParsedModel = GeminiModel | AnthropicModel | OpenAIModel | UnknownMo
 /** Strip a provider namespace prefix (`openai/gpt-5.4` → `gpt-5.4`). */
 export declare function bareModelId(modelId: string): string;
 export declare function parseKnownModel(modelId: string): ParsedModel;
-export declare function parseGeminiModel(modelId: string): GeminiModel | null;
-export declare function parseAnthropicModel(modelId: string): AnthropicModel | null;
-export declare function parseOpenAIModel(modelId: string): OpenAIModel | null;
+export declare const parseGeminiModel: (modelId: string) => GeminiModel | null;
+export declare const parseAnthropicModel: (modelId: string) => AnthropicModel | null;
+export declare const parseOpenAIModel: (modelId: string) => OpenAIModel | null;
+/**
+ * Parse a GLM (Zhipu / Z.AI) model id into family + variant + vision + version.
+ * Shape: `glm-<version>[v][-<variant>]` — e.g. `glm-4.5`, `glm-4.5-air`,
+ * `glm-5-turbo`, `glm-4.5v`, `glm-5-preview`. The `v` (vision) attaches to the
+ * version; other variants are `-` suffixes. Standalone like `parseAnthropicModel`
+ * is used in family.ts — GLM needs no global thinking policy, so it stays out of
+ * `parseKnownModel`.
+ */
+export declare const parseGlmModel: (modelId: string) => GlmModel | null;
 export declare function isFableOrMythos(kind: AnthropicKind): boolean;
 export declare function parseSemVer(version: string): SemVer | null;
 export declare function semverGte(left: SemVer | string, right: SemVer | string): boolean;

package/dist/types/identity/family.d.ts

@@ -37,6 +37,28 @@ export declare function isMinimaxM2FamilyModelId(modelId: string): boolean;
  * and `none`.
  */
 export declare function isOpenAIGptOssModelId(modelId: string): boolean;
+/**
+ * Reasoning-capable GLM coding SKUs: glm-4.5 and up on the base / `-air` /
+ * `-turbo` lines. Excludes the vision (`…v`) shape, the non-reasoning
+ * `-flash`/`-flashx`/`-preview` variants, and pre-4.5 ids. Matching the family
+ * keeps newly-bumped integers (`glm-5.3`, `glm-6`, …) covered without a per-id
+ * allowlist.
+ */
+export declare function isReasoningGlmModelId(modelId: string): boolean;
+/** GLM vision SKUs — the `v` that attaches to the version (`glm-4v`, `glm-4.5v`). */
+export declare function isGlmVisionModelId(modelId: string): boolean;
+/**
+ * Coarse vendor-lineage token for "are two models the same family?" checks
+ * (e.g. picking a cross-family reviewer). All Claude point releases share a token,
+ * Claude and GPT differ; namespace prefixes and aggregator mirrors fold onto the
+ * lineage via {@link parseKnownModel}'s `bareModelId` normalization. Opaque and
+ * comparison-only — not a stable key to persist, since the vocabulary tracks new
+ * releases. Returns `""` for ids it cannot classify; callers fall back to the provider.
+ *
+ * Vendor-only by design: a model's kind/variant (opus vs sonnet, codex vs base) is
+ * collapsed onto the single vendor token; use {@link parseKnownModel} for finer breakdowns.
+ */
+export declare function modelFamilyToken(modelId: string): string;
 /**
  * Adaptive thinking `display` is supported starting with Claude Opus 4.7 and
  * the Claude Fable/Mythos 5 generation. Older adaptive-thinking models

package/dist/types/model-manager.d.ts

@@ -22,6 +22,8 @@ export interface ModelManagerOptions<TApi extends Api = Api, TModelsDevPayload =
     staticModels?: readonly ModelSpec<TApi>[];
     /** Optional override for the cache database path. Default: <agent-dir>/models.db. */
     cacheDbPath?: string;
+    /** Optional provider id override for cache namespacing. Defaults to providerId. */
+    cacheProviderId?: string;
     /** Maximum cache age in milliseconds before considered stale. Default: 24h. */
     cacheTtlMs?: number;
     /** When true, a successful dynamic fetch is the complete provider catalog and prunes static-only models. */

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

@@ -25,10 +25,10 @@ export declare const CATALOG_PROVIDERS: readonly [{
     };
 }, {
     readonly id: "amazon-bedrock";
-    readonly defaultModel: "us.anthropic.claude-opus-4-6-v1";
+    readonly defaultModel: "us.anthropic.claude-opus-4-8";
 }, {
     readonly id: "anthropic";
-    readonly defaultModel: "claude-opus-4-6";
+    readonly defaultModel: "claude-opus-4-8";
     readonly createModelManagerOptions: (config: ModelManagerConfig) => import("..").ModelManagerOptions<"anthropic-messages", unknown>;
 }, {
     readonly id: "cerebras";
@@ -136,7 +136,7 @@ export declare const CATALOG_PROVIDERS: readonly [{
     };
 }, {
     readonly id: "litellm";
-    readonly defaultModel: "claude-opus-4-6";
+    readonly defaultModel: "claude-opus-4-8";
     readonly envVars: readonly ["LITELLM_API_KEY"];
     readonly createModelManagerOptions: (config: ModelManagerConfig) => import("..").ModelManagerOptions<"openai-completions", unknown>;
     readonly catalogDiscovery: {
@@ -176,7 +176,7 @@ export declare const CATALOG_PROVIDERS: readonly [{
     };
 }, {
     readonly id: "nanogpt";
-    readonly defaultModel: "openai/gpt-5.4";
+    readonly defaultModel: "openai/gpt-5.5";
     readonly envVars: readonly ["NANO_GPT_API_KEY"];
     readonly createModelManagerOptions: (config: ModelManagerConfig) => import("..").ModelManagerOptions<"openai-completions", unknown>;
     readonly catalogDiscovery: {
@@ -207,12 +207,12 @@ export declare const CATALOG_PROVIDERS: readonly [{
     };
 }, {
     readonly id: "openai";
-    readonly defaultModel: "gpt-5.4";
+    readonly defaultModel: "gpt-5.5";
     readonly envVars: readonly ["OPENAI_API_KEY"];
     readonly createModelManagerOptions: (config: ModelManagerConfig) => import("..").ModelManagerOptions<"openai-responses", unknown>;
 }, {
     readonly id: "openai-codex";
-    readonly defaultModel: "gpt-5.4";
+    readonly defaultModel: "gpt-5.5";
     readonly envVars: readonly ["OPENAI_CODEX_OAUTH_TOKEN"];
     readonly specialModelManager: true;
 }, {
@@ -227,7 +227,7 @@ export declare const CATALOG_PROVIDERS: readonly [{
     readonly createModelManagerOptions: (config: ModelManagerConfig) => import("..").ModelManagerOptions<import("..").Api, unknown>;
 }, {
     readonly id: "openrouter";
-    readonly defaultModel: "openai/gpt-5.4";
+    readonly defaultModel: "openai/gpt-5.5";
     readonly envVars: readonly ["OPENROUTER_API_KEY"];
     readonly createModelManagerOptions: (config: ModelManagerConfig) => import("..").ModelManagerOptions<"openai-completions", unknown>;
     readonly catalogDiscovery: {
@@ -361,7 +361,7 @@ export declare const CATALOG_PROVIDERS: readonly [{
     };
 }, {
     readonly id: "zenmux";
-    readonly defaultModel: "anthropic/claude-opus-4.6";
+    readonly defaultModel: "anthropic/claude-opus-4.8";
     readonly envVars: readonly ["ZENMUX_API_KEY"];
     readonly createModelManagerOptions: (config: ModelManagerConfig) => import("..").ModelManagerOptions<import("..").Api, unknown>;
     readonly catalogDiscovery: {

package/dist/types/types.d.ts

@@ -161,6 +161,12 @@ export interface OpenAICompat {
     requiresAssistantContentForToolCalls?: boolean;
     /** Whether the provider supports the `tool_choice` parameter. Default: true. */
     supportsToolChoice?: boolean;
+    /**
+     * Whether forced `tool_choice` values (`"required"` or named tools) are accepted.
+     * When false, request builders keep tools available but downgrade forced choices
+     * to provider-default auto selection. Default: true.
+     */
+    supportsForcedToolChoice?: boolean;
     /**
      * Drop reasoning fields (`reasoning_effort`, OpenRouter `reasoning`) for
      * the request when `tool_choice` forces a tool call. Mirrors the Anthropic

package/dist/types/wire/gemini-headers.d.ts

@@ -9,6 +9,7 @@ export declare const getGeminiCliHeaders: (modelId?: string) => {
     "Client-Metadata": string;
 };
 export declare const ANTIGRAVITY_SYSTEM_INSTRUCTION: string;
+export declare const ANTIGRAVITY_NO_PREAMBLE_INSTRUCTION = "CRITICAL: NEVER output rule checks, formatting guidelines, constraint checklists (e.g. \"No emdashes\"), or your thinking/personality preambles in the final response. Output only the final response.";
 /**
  * Antigravity / Cloud Code Assist user agent. Lives in its own file so discovery
  * and usage code can read it without pulling the heavy google-gemini-cli provider

package/package.json

@@ -1,7 +1,7 @@
 {
 	"type": "module",
 	"name": "@oh-my-pi/pi-catalog",
-	"version": "15.12.4",
+	"version": "15.13.1",
 	"description": "Model catalog for omp: bundled model database, provider discovery descriptors, model identity, classification, and equivalence",
 	"homepage": "https://omp.sh",
 	"author": "Can Boluk",
@@ -34,11 +34,11 @@
 	},
 	"dependencies": {
 		"@bufbuild/protobuf": "^2.12.0",
-		"@oh-my-pi/pi-utils": "15.12.4",
+		"@oh-my-pi/pi-utils": "15.13.1",
 		"zod": "^4"
 	},
 	"devDependencies": {
-		"@oh-my-pi/pi-ai": "15.12.4",
+		"@oh-my-pi/pi-ai": "15.13.1",
 		"@types/bun": "^1.3.14"
 	},
 	"engines": {

package/src/compat/anthropic.ts

@@ -34,11 +34,17 @@ export function buildAnthropicCompat(spec: ModelSpec<"anthropic-messages">): Res
 	const official = isOfficialAnthropicApiUrl(baseUrl);
 	// Z.AI's Anthropic-compatible proxy lives at `api.z.ai/api/anthropic`.
 	const isZai = modelMatchesHost(spec, "zai");
+	// GitHub Copilot's Anthropic-compatible proxy (api.githubcopilot.com/v1/messages)
+	// rejects the per-tool `eager_input_streaming` field with
+	// `tools.0.custom.eager_input_streaming: Extra inputs are not permitted` and
+	// doesn't whitelist the `fine-grained-tool-streaming-2025-05-14` beta either
+	// (issue #2558), so eager tool-input streaming is unavailable on this host.
+	const isCopilot = modelMatchesHost(spec, "githubCopilot");
 	const compat: ResolvedAnthropicCompat = {
 		officialEndpoint: official,
 		disableStrictTools: false,
 		disableAdaptiveThinking: false,
-		supportsEagerToolInputStreaming: true,
+		supportsEagerToolInputStreaming: !isCopilot,
 		// Long cache retention is only sent to the official API by default;
 		// proxies opt in explicitly via `compat.supportsLongCacheRetention: true`.
 		supportsLongCacheRetention: official,

package/src/compat/openai.ts

@@ -217,6 +217,7 @@ export function buildOpenAICompat(spec: ModelSpec<"openai-completions">): Resolv
 		disableReasoningOnForcedToolChoice: isKimiModel || isAnthropicModel,
 		disableReasoningOnToolChoice: isDeepseekFamily && Boolean(spec.reasoning) && !isOpenRouter,
 		supportsToolChoice: !isDirectDeepseekReasoning,
+		supportsForcedToolChoice: true,
 		maxTokensField: useMaxTokens ? "max_tokens" : "max_completion_tokens",
 		requiresToolResultName: isMistral,
 		requiresAssistantAfterToolResult: false,

package/src/identity/classify.ts

@@ -14,6 +14,7 @@ export type SemVer = {
 export type GeminiKind = "pro" | "flash";
 export type AnthropicKind = "opus" | "sonnet" | "fable" | "mythos";
 export type OpenAIVariant = "base" | "codex" | "codex-max" | "codex-mini" | "codex-spark" | "mini" | "max" | "nano";
+export type GlmVariant = "base" | "air" | "turbo" | "flash" | "flashx" | "preview";
 
 export interface GeminiModel {
 	family: "gemini";
@@ -33,6 +34,15 @@ export interface OpenAIModel {
 	version: SemVer;
 }
 
+export interface GlmModel {
+	family: "glm";
+	/** Suffix variant (`-air`, `-turbo`, `-flash`, `-flashx`, `-preview`); `base` when none. */
+	variant: GlmVariant;
+	/** Vision SKU — the `v` that attaches directly to the version (`glm-4v`, `glm-4.5v`). */
+	vision: boolean;
+	version: SemVer;
+}
+
 export interface UnknownModel {
 	family: "unknown";
 	id: string;
@@ -55,8 +65,26 @@ export function parseKnownModel(modelId: string): ParsedModel {
 	);
 }
 
+/**
+ * Wrap a parse function in a per-id memo cache. Caches the `null` result too, so
+ * repeated misses (the common case — ids of other families) stay O(1) and never
+ * re-run the regex/semver work.
+ */
+function parser<T>(parse: (modelId: string) => T | null): (modelId: string) => T | null {
+	const cache = new Map<string, T | null>();
+	return modelId => {
+		const hit = cache.get(modelId);
+		if (hit !== undefined || cache.has(modelId)) {
+			return hit ?? null;
+		}
+		const result = parse(modelId);
+		cache.set(modelId, result);
+		return result;
+	};
+}
+
 const GEMINI_SUFFIX = "-preview";
-export function parseGeminiModel(modelId: string): GeminiModel | null {
+export const parseGeminiModel = parser((modelId): GeminiModel | null => {
 	if (modelId.endsWith(GEMINI_SUFFIX)) {
 		modelId = modelId.slice(0, -GEMINI_SUFFIX.length);
 	}
@@ -69,9 +97,9 @@ export function parseGeminiModel(modelId: string): GeminiModel | null {
 		return null;
 	}
 	return { family: "gemini", kind: match[2] as GeminiKind, version };
-}
+});
 
-export function parseAnthropicModel(modelId: string): AnthropicModel | null {
+export const parseAnthropicModel = parser((modelId): AnthropicModel | null => {
 	const match = /claude-(opus|sonnet|fable|mythos)-(\d{1,2}(?:[.-]\d{1,2}){0,2})\b/.exec(modelId);
 	if (!match) {
 		return null;
@@ -81,9 +109,9 @@ export function parseAnthropicModel(modelId: string): AnthropicModel | null {
 		return null;
 	}
 	return { family: "anthropic", kind: match[1] as AnthropicKind, version };
-}
+});
 
-export function parseOpenAIModel(modelId: string): OpenAIModel | null {
+export const parseOpenAIModel = parser((modelId): OpenAIModel | null => {
 	const match = /gpt-(\d+(?:\.\d+){0,2})(?:-(codex-spark|codex-mini|codex-max|codex|mini|max|nano))?\b/.exec(modelId);
 	if (!match) {
 		return null;
@@ -93,7 +121,32 @@ export function parseOpenAIModel(modelId: string): OpenAIModel | null {
 		return null;
 	}
 	return { family: "openai", variant: (match[2] as OpenAIVariant | undefined) ?? "base", version };
-}
+});
+
+/**
+ * Parse a GLM (Zhipu / Z.AI) model id into family + variant + vision + version.
+ * Shape: `glm-<version>[v][-<variant>]` — e.g. `glm-4.5`, `glm-4.5-air`,
+ * `glm-5-turbo`, `glm-4.5v`, `glm-5-preview`. The `v` (vision) attaches to the
+ * version; other variants are `-` suffixes. Standalone like `parseAnthropicModel`
+ * is used in family.ts — GLM needs no global thinking policy, so it stays out of
+ * `parseKnownModel`.
+ */
+export const parseGlmModel = parser((modelId): GlmModel | null => {
+	const match = /glm-(\d{1,2}(?:\.\d+)?)(v)?(?:-(air|turbo|flashx|flash|preview))?\b/.exec(modelId);
+	if (!match) {
+		return null;
+	}
+	const version = parseSemVer(match[1]);
+	if (!version) {
+		return null;
+	}
+	return {
+		family: "glm",
+		variant: (match[3] as GlmVariant | undefined) ?? "base",
+		vision: match[2] === "v",
+		version,
+	};
+});
 
 export function isFableOrMythos(kind: AnthropicKind): boolean {
 	return kind === "fable" || kind === "mythos";

package/src/identity/family.ts

@@ -7,7 +7,14 @@
  * here.
  */
 
-import { bareModelId, isFableOrMythos, parseAnthropicModel, semverGte } from "./classify";
+import {
+	bareModelId,
+	isFableOrMythos,
+	parseAnthropicModel,
+	parseGlmModel,
+	parseKnownModel,
+	semverGte,
+} from "./classify";
 
 /** Kimi family ids in any namespace form (`moonshotai/kimi-*`, `kimi-k2.6`, `vendor/kimi.x`). */
 export function isKimiModelId(modelId: string): boolean {
@@ -71,6 +78,52 @@ export function isOpenAIGptOssModelId(modelId: string): boolean {
 	return /(^|\/)gpt-oss[-:]/i.test(modelId);
 }
 
+/**
+ * Reasoning-capable GLM coding SKUs: glm-4.5 and up on the base / `-air` /
+ * `-turbo` lines. Excludes the vision (`…v`) shape, the non-reasoning
+ * `-flash`/`-flashx`/`-preview` variants, and pre-4.5 ids. Matching the family
+ * keeps newly-bumped integers (`glm-5.3`, `glm-6`, …) covered without a per-id
+ * allowlist.
+ */
+export function isReasoningGlmModelId(modelId: string): boolean {
+	const glm = parseGlmModel(bareModelId(modelId));
+	if (!glm || glm.vision) {
+		return false;
+	}
+	if (glm.variant !== "base" && glm.variant !== "air" && glm.variant !== "turbo") {
+		return false;
+	}
+	return semverGte(glm.version, "4.5");
+}
+
+/** GLM vision SKUs — the `v` that attaches to the version (`glm-4v`, `glm-4.5v`). */
+export function isGlmVisionModelId(modelId: string): boolean {
+	return parseGlmModel(bareModelId(modelId))?.vision === true;
+}
+/**
+ * Coarse vendor-lineage token for "are two models the same family?" checks
+ * (e.g. picking a cross-family reviewer). All Claude point releases share a token,
+ * Claude and GPT differ; namespace prefixes and aggregator mirrors fold onto the
+ * lineage via {@link parseKnownModel}'s `bareModelId` normalization. Opaque and
+ * comparison-only — not a stable key to persist, since the vocabulary tracks new
+ * releases. Returns `""` for ids it cannot classify; callers fall back to the provider.
+ *
+ * Vendor-only by design: a model's kind/variant (opus vs sonnet, codex vs base) is
+ * collapsed onto the single vendor token; use {@link parseKnownModel} for finer breakdowns.
+ */
+export function modelFamilyToken(modelId: string): string {
+	const parsed = parseKnownModel(modelId);
+	if (parsed.family !== "unknown") return parsed.family;
+	if (isKimiModelId(modelId)) return "kimi";
+	if (isQwenModelId(modelId)) return "qwen";
+	if (isMinimaxM2FamilyModelId(modelId)) return "minimax";
+	if (isOpenAIGptOssModelId(modelId)) return "gpt-oss";
+	if (isDeepseekModelIdOrName(modelId)) return "deepseek";
+	if (isMimoModelIdOrName(modelId)) return "mimo";
+	if (parseGlmModel(bareModelId(modelId))) return "glm";
+	return "";
+}
+
 /**
  * Adaptive thinking `display` is supported starting with Claude Opus 4.7 and
  * the Claude Fable/Mythos 5 generation. Older adaptive-thinking models

package/src/model-manager.ts

@@ -33,6 +33,8 @@ export interface ModelManagerOptions<TApi extends Api = Api, TModelsDevPayload =
 	staticModels?: readonly ModelSpec<TApi>[];
 	/** Optional override for the cache database path. Default: <agent-dir>/models.db. */
 	cacheDbPath?: string;
+	/** Optional provider id override for cache namespacing. Defaults to providerId. */
+	cacheProviderId?: string;
 	/** Maximum cache age in milliseconds before considered stale. Default: 24h. */
 	cacheTtlMs?: number;
 	/** When true, a successful dynamic fetch is the complete provider catalog and prunes static-only models. */
@@ -107,13 +109,14 @@ export async function resolveProviderModels<TApi extends Api = Api, TModelsDevPa
 	options: ModelManagerOptions<TApi, TModelsDevPayload>,
 	strategy: ModelRefreshStrategy = "online-if-uncached",
 ): Promise<ModelResolutionResult<TApi>> {
+	const cacheProviderId = options.cacheProviderId ?? options.providerId;
 	const now = options.now ?? Date.now;
 	const ttlMs = options.cacheTtlMs ?? DEFAULT_CACHE_TTL_MS;
 	const dbPath = options.cacheDbPath;
 	const staticModels = options.staticModels
 		? passModelList<TApi>(options.staticModels)
 		: (getBundledModels(options.providerId as GeneratedProvider) as Model<TApi>[]);
-	const cache = readModelCache<TApi>(options.providerId, ttlMs, now, dbPath);
+	const cache = readModelCache<TApi>(cacheProviderId, ttlMs, now, dbPath);
 	const dynamicModelsAuthoritative = options.dynamicModelsAuthoritative ?? false;
 	const staticFingerprint = fingerprintStatic(staticModels, dynamicModelsAuthoritative);
 	const cacheFingerprintMatches = cache?.staticFingerprint === staticFingerprint && staticFingerprint.length > 0;
@@ -160,7 +163,7 @@ export async function resolveProviderModels<TApi extends Api = Api, TModelsDevPa
 				? retainModelIds(mergedSnapshot, dynamicModels)
 				: mergedSnapshot;
 			writeModelCache(
-				options.providerId,
+				cacheProviderId,
 				now(),
 				collapseBuiltModelVariants(snapshotModels),
 				true,
@@ -170,9 +173,9 @@ export async function resolveProviderModels<TApi extends Api = Api, TModelsDevPa
 		} else {
 			// Dynamic fetch failed — update cache with a non-authoritative snapshot so
 			// stale state remains visible while retry backoff still applies.
-			const latestCache = readModelCache<TApi>(options.providerId, ttlMs, now, dbPath);
+			const latestCache = readModelCache<TApi>(cacheProviderId, ttlMs, now, dbPath);
 			writeModelCache(
-				options.providerId,
+				cacheProviderId,
 				now(),
 				collapseBuiltModelVariants(
 					mergeDynamicModels(

package/src/models.json

@@ -4259,7 +4259,8 @@
 				"cacheWrite": 0
 			},
 			"contextWindow": null,
-			"maxTokens": null
+			"maxTokens": null,
+			"contextPromotionTarget": "aimlapi/gpt-5.4-2026-03-05"
 		},
 		"gpt-5.5-pro-2026-04-23": {
 			"id": "gpt-5.5-pro-2026-04-23",
@@ -4278,7 +4279,8 @@
 				"cacheWrite": 0
 			},
 			"contextWindow": null,
-			"maxTokens": null
+			"maxTokens": null,
+			"contextPromotionTarget": "aimlapi/gpt-5.4-2026-03-05"
 		},
 		"gpt-oss-120b": {
 			"id": "gpt-oss-120b",
@@ -9577,7 +9579,8 @@
 					"high",
 					"xhigh"
 				]
-			}
+			},
+			"contextPromotionTarget": "amazon-bedrock/openai.gpt-5.4"
 		},
 		"openai.gpt-oss-120b": {
 			"id": "openai.gpt-oss-120b",
@@ -12202,7 +12205,8 @@
 					"high",
 					"xhigh"
 				]
-			}
+			},
+			"contextPromotionTarget": "cloudflare-ai-gateway/openai/gpt-5.4"
 		},
 		"openai/o1": {
 			"id": "openai/o1",
@@ -22904,6 +22908,7 @@
 				"disableReasoningOnForcedToolChoice": true,
 				"disableReasoningOnToolChoice": false,
 				"supportsToolChoice": true,
+				"supportsForcedToolChoice": true,
 				"maxTokensField": "max_completion_tokens",
 				"requiresToolResultName": false,
 				"requiresAssistantAfterToolResult": false,
@@ -24595,7 +24600,8 @@
 					"high",
 					"xhigh"
 				]
-			}
+			},
+			"contextPromotionTarget": "kilo/openai/gpt-5.4"
 		},
 		"openai/gpt-5.5-pro": {
 			"id": "openai/gpt-5.5-pro",
@@ -24624,7 +24630,8 @@
 					"high",
 					"xhigh"
 				]
-			}
+			},
+			"contextPromotionTarget": "kilo/openai/gpt-5.4"
 		},
 		"openai/gpt-audio": {
 			"id": "openai/gpt-audio",
@@ -25327,6 +25334,7 @@
 				"disableReasoningOnForcedToolChoice": false,
 				"disableReasoningOnToolChoice": false,
 				"supportsToolChoice": true,
+				"supportsForcedToolChoice": true,
 				"maxTokensField": "max_completion_tokens",
 				"requiresToolResultName": false,
 				"requiresAssistantAfterToolResult": false,
@@ -25778,6 +25786,7 @@
 				"disableReasoningOnForcedToolChoice": false,
 				"disableReasoningOnToolChoice": false,
 				"supportsToolChoice": true,
+				"supportsForcedToolChoice": true,
 				"maxTokensField": "max_completion_tokens",
 				"requiresToolResultName": false,
 				"requiresAssistantAfterToolResult": false,
@@ -26058,6 +26067,7 @@
 				"disableReasoningOnForcedToolChoice": false,
 				"disableReasoningOnToolChoice": false,
 				"supportsToolChoice": true,
+				"supportsForcedToolChoice": true,
 				"maxTokensField": "max_completion_tokens",
 				"requiresToolResultName": false,
 				"requiresAssistantAfterToolResult": false,
@@ -28539,7 +28549,7 @@
 				"cacheRead": 0.12,
 				"cacheWrite": 0
 			},
-			"contextWindow": 512000,
+			"contextWindow": 1000000,
 			"maxTokens": 128000,
 			"thinking": {
 				"mode": "budget",
@@ -28781,7 +28791,7 @@
 				"cacheRead": 0.12,
 				"cacheWrite": 0
 			},
-			"contextWindow": 512000,
+			"contextWindow": 1000000,
 			"maxTokens": 128000,
 			"thinking": {
 				"mode": "budget",
@@ -39194,8 +39204,8 @@
 				"cacheRead": 0,
 				"cacheWrite": 0
 			},
-			"contextWindow": null,
-			"maxTokens": null
+			"contextWindow": 128000,
+			"maxTokens": 16384
 		},
 		"openai/gpt-5-codex": {
 			"id": "openai/gpt-5-codex",
@@ -39763,7 +39773,8 @@
 					"high",
 					"xhigh"
 				]
-			}
+			},
+			"contextPromotionTarget": "nanogpt/openai/gpt-5.4"
 		},
 		"openai/gpt-chat-latest": {
 			"id": "openai/gpt-chat-latest",
@@ -51042,6 +51053,9 @@
 			},
 			"contextWindow": 262144,
 			"maxTokens": 262144,
+			"compat": {
+				"supportsForcedToolChoice": false
+			},
 			"thinking": {
 				"mode": "effort",
 				"efforts": [
@@ -51081,6 +51095,9 @@
 					"high",
 					"xhigh"
 				]
+			},
+			"compat": {
+				"supportsToolChoice": false
 			}
 		},
 		"mimo-v2-pro": {
@@ -51110,6 +51127,9 @@
 					"high",
 					"xhigh"
 				]
+			},
+			"compat": {
+				"supportsToolChoice": false
 			}
 		},
 		"mimo-v2.5": {
@@ -51131,6 +51151,9 @@
 			},
 			"contextWindow": 1000000,
 			"maxTokens": 128000,
+			"compat": {
+				"supportsToolChoice": false
+			},
 			"thinking": {
 				"mode": "effort",
 				"efforts": [
@@ -51160,6 +51183,9 @@
 			},
 			"contextWindow": 1048576,
 			"maxTokens": 128000,
+			"compat": {
+				"supportsToolChoice": false
+			},
 			"thinking": {
 				"mode": "effort",
 				"efforts": [
@@ -55012,13 +55038,13 @@
 				"text"
 			],
 			"cost": {
-				"input": 0.098,
-				"output": 0.196,
+				"input": 0.09,
+				"output": 0.18,
 				"cacheRead": 0.02,
 				"cacheWrite": 0
 			},
 			"contextWindow": 1048576,
-			"maxTokens": 384000,
+			"maxTokens": 65536,
 			"thinking": {
 				"mode": "effort",
 				"efforts": [
@@ -57075,9 +57101,9 @@
 				"image"
 			],
 			"cost": {
-				"input": 0.95,
-				"output": 4,
-				"cacheRead": 0.19,
+				"input": 0.75,
+				"output": 3.5,
+				"cacheRead": 0.16,
 				"cacheWrite": 0
 			},
 			"contextWindow": 262144,
@@ -58513,7 +58539,8 @@
 					"high",
 					"xhigh"
 				]
-			}
+			},
+			"contextPromotionTarget": "openrouter/openai/gpt-5.4"
 		},
 		"openai/gpt-5.5-pro": {
 			"id": "openai/gpt-5.5-pro",
@@ -58542,7 +58569,8 @@
 					"high",
 					"xhigh"
 				]
-			}
+			},
+			"contextPromotionTarget": "openrouter/openai/gpt-5.4"
 		},
 		"openai/gpt-audio": {
 			"id": "openai/gpt-audio",
@@ -59989,7 +60017,7 @@
 				"cacheWrite": 0
 			},
 			"contextWindow": 262144,
-			"maxTokens": null
+			"maxTokens": 16384
 		},
 		"qwen/qwen3-next-80b-a3b-thinking": {
 			"id": "qwen/qwen3-next-80b-a3b-thinking",
@@ -64583,7 +64611,7 @@
 				"cacheWrite": 0
 			},
 			"contextWindow": 128000,
-			"maxTokens": null,
+			"maxTokens": 16384,
 			"compat": {
 				"supportsUsageInStreaming": false
 			}
@@ -69051,7 +69079,8 @@
 					"high",
 					"xhigh"
 				]
-			}
+			},
+			"contextPromotionTarget": "vercel-ai-gateway/openai/gpt-5.4"
 		},
 		"openai/gpt-5.5-pro": {
 			"id": "openai/gpt-5.5-pro",
@@ -69080,7 +69109,8 @@
 					"high",
 					"xhigh"
 				]
-			}
+			},
+			"contextPromotionTarget": "vercel-ai-gateway/openai/gpt-5.4"
 		},
 		"openai/gpt-oss-120b": {
 			"id": "openai/gpt-oss-120b",
@@ -72205,6 +72235,35 @@
 				]
 			}
 		},
+		"glm-5.2": {
+			"id": "glm-5.2",
+			"name": "GLM-5.2",
+			"api": "anthropic-messages",
+			"provider": "zai",
+			"baseUrl": "https://api.z.ai/api/anthropic",
+			"reasoning": true,
+			"input": [
+				"text"
+			],
+			"cost": {
+				"input": 0,
+				"output": 0,
+				"cacheRead": 0,
+				"cacheWrite": 0
+			},
+			"contextWindow": 1000000,
+			"maxTokens": 131072,
+			"thinking": {
+				"mode": "budget",
+				"efforts": [
+					"minimal",
+					"low",
+					"medium",
+					"high",
+					"xhigh"
+				]
+			}
+		},
 		"glm-5v-turbo": {
 			"id": "glm-5v-turbo",
 			"name": "GLM-5V-Turbo",
@@ -75112,7 +75171,8 @@
 					"high",
 					"xhigh"
 				]
-			}
+			},
+			"contextPromotionTarget": "zenmux/openai/gpt-5.4"
 		},
 		"openai/gpt-5.5-instant": {
 			"id": "openai/gpt-5.5-instant",
@@ -75141,7 +75201,8 @@
 					"high",
 					"xhigh"
 				]
-			}
+			},
+			"contextPromotionTarget": "zenmux/openai/gpt-5.4"
 		},
 		"openai/gpt-5.5-pro": {
 			"id": "openai/gpt-5.5-pro",
@@ -75170,7 +75231,8 @@
 					"high",
 					"xhigh"
 				]
-			}
+			},
+			"contextPromotionTarget": "zenmux/openai/gpt-5.4"
 		},
 		"openai/gpt-image-1.5": {
 			"id": "openai/gpt-image-1.5",

package/src/provider-models/descriptors.ts

@@ -68,11 +68,11 @@ export const CATALOG_PROVIDERS = [
 	},
 	{
 		id: "amazon-bedrock",
-		defaultModel: "us.anthropic.claude-opus-4-6-v1",
+		defaultModel: "us.anthropic.claude-opus-4-8",
 	},
 	{
 		id: "anthropic",
-		defaultModel: "claude-opus-4-6",
+		defaultModel: "claude-opus-4-8",
 		createModelManagerOptions: (config: ModelManagerConfig) => anthropicModelManagerOptions(config),
 	},
 	{
@@ -177,7 +177,7 @@ export const CATALOG_PROVIDERS = [
 	},
 	{
 		id: "litellm",
-		defaultModel: "claude-opus-4-6",
+		defaultModel: "claude-opus-4-8",
 		envVars: ["LITELLM_API_KEY"],
 		createModelManagerOptions: (config: ModelManagerConfig) => litellmModelManagerOptions(config),
 		catalogDiscovery: { label: "LiteLLM", allowUnauthenticated: true },
@@ -219,7 +219,7 @@ export const CATALOG_PROVIDERS = [
 	},
 	{
 		id: "nanogpt",
-		defaultModel: "openai/gpt-5.4",
+		defaultModel: "openai/gpt-5.5",
 		envVars: ["NANO_GPT_API_KEY"],
 		createModelManagerOptions: (config: ModelManagerConfig) => nanoGptModelManagerOptions(config),
 		catalogDiscovery: { label: "NanoGPT" },
@@ -247,13 +247,13 @@ export const CATALOG_PROVIDERS = [
 	},
 	{
 		id: "openai",
-		defaultModel: "gpt-5.4",
+		defaultModel: "gpt-5.5",
 		envVars: ["OPENAI_API_KEY"],
 		createModelManagerOptions: (config: ModelManagerConfig) => openaiModelManagerOptions(config),
 	},
 	{
 		id: "openai-codex",
-		defaultModel: "gpt-5.4",
+		defaultModel: "gpt-5.5",
 		envVars: ["OPENAI_CODEX_OAUTH_TOKEN"],
 		specialModelManager: true,
 	},
@@ -271,7 +271,7 @@ export const CATALOG_PROVIDERS = [
 	},
 	{
 		id: "openrouter",
-		defaultModel: "openai/gpt-5.4",
+		defaultModel: "openai/gpt-5.5",
 		envVars: ["OPENROUTER_API_KEY"],
 		createModelManagerOptions: (config: ModelManagerConfig) => openrouterModelManagerOptions(config),
 		catalogDiscovery: { label: "OpenRouter", allowUnauthenticated: true },
@@ -403,7 +403,7 @@ export const CATALOG_PROVIDERS = [
 	},
 	{
 		id: "zenmux",
-		defaultModel: "anthropic/claude-opus-4.6",
+		defaultModel: "anthropic/claude-opus-4.8",
 		envVars: ["ZENMUX_API_KEY"],
 		createModelManagerOptions: (config: ModelManagerConfig) => zenmuxModelManagerOptions(config),
 		catalogDiscovery: { label: "ZenMux" },

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

@@ -5,6 +5,7 @@ import {
 } from "../discovery/openai-compatible";
 import { Effort } from "../effort";
 import { toFireworksPublicModelId } from "../fireworks-model-id";
+import { isGlmVisionModelId, isReasoningGlmModelId } from "../identity/family";
 import type { ModelManagerOptions } from "../model-manager";
 import { getBundledModels } from "../models";
 import type { Api, FetchImpl, Model, ModelSpec, Provider, ThinkingConfig } from "../types";
@@ -1030,8 +1031,8 @@ export function zhipuCodingPlanModelManagerOptions(
 						const id = defaults.id;
 						return {
 							...defaults,
-							reasoning: ZHIPU_REASONING_MODELS[id] === true || id.includes("thinking"),
-							input: ZHIPU_VISION_PATTERN.test(id) ? (["text", "image"] as const) : ["text"],
+							reasoning: isReasoningGlmModelId(id) || id.includes("thinking"),
+							input: isGlmVisionModelId(id) ? (["text", "image"] as const) : ["text"],
 							compat: {
 								thinkingFormat: "zai",
 								reasoningContentField: "reasoning_content",
@@ -1045,25 +1046,6 @@ export function zhipuCodingPlanModelManagerOptions(
 	};
 }
 
-// Reasoning-capable GLM models on the BigModel coding-plan SKU. Keep this
-// explicit rather than regex-matching `glm-[45]\.\d` so newly-added integers
-// like `glm-5` / `glm-5-turbo` are covered and unrelated future SKUs (e.g.
-// `glm-5-preview`) do not silently flip into thinking mode.
-const ZHIPU_REASONING_MODELS: Readonly<Record<string, true>> = {
-	"glm-4.5": true,
-	"glm-4.5-air": true,
-	"glm-4.6": true,
-	"glm-4.7": true,
-	"glm-5": true,
-	"glm-5-turbo": true,
-	"glm-5.1": true,
-};
-
-// Vision-capable GLM models follow the `glm-<N>[.<N>]v[-<variant>]` shape
-// (e.g. `glm-4v`, `glm-4.5v`, `glm-4v-plus`). The previous `id.includes("v")`
-// check matched anything with a `v` — including the non-vision `glm-5-preview`.
-const ZHIPU_VISION_PATTERN = /^glm-[45](?:\.\d+)?v(?:-|$)/;
-
 // ---------------------------------------------------------------------------
 // 7.5 Fireworks
 // ---------------------------------------------------------------------------
@@ -2393,6 +2375,8 @@ export function litellmModelManagerOptions(
 // 22. vLLM
 // ---------------------------------------------------------------------------
 
+const VLLM_DISCOVERY_TIMEOUT_MS = 10_000;
+
 export interface VllmModelManagerConfig {
 	apiKey?: string;
 	baseUrl?: string;
@@ -2405,6 +2389,7 @@ export function vllmModelManagerOptions(config?: VllmModelManagerConfig): ModelM
 	const references = createBundledReferenceMap<"openai-completions">("vllm" as Parameters<typeof getBundledModels>[0]);
 	return {
 		providerId: "vllm",
+		cacheProviderId: `vllm:${Bun.hash(baseUrl).toString(36)}`,
 		fetchDynamicModels: () =>
 			fetchOpenAICompatibleModels({
 				api: "openai-completions",
@@ -2419,6 +2404,7 @@ export function vllmModelManagerOptions(config?: VllmModelManagerConfig): ModelM
 					};
 				},
 				fetch: config?.fetch,
+				signal: AbortSignal.timeout(VLLM_DISCOVERY_TIMEOUT_MS),
 			}),
 	};
 }

package/src/types.ts

@@ -186,6 +186,12 @@ export interface OpenAICompat {
 	requiresAssistantContentForToolCalls?: boolean;
 	/** Whether the provider supports the `tool_choice` parameter. Default: true. */
 	supportsToolChoice?: boolean;
+	/**
+	 * Whether forced `tool_choice` values (`"required"` or named tools) are accepted.
+	 * When false, request builders keep tools available but downgrade forced choices
+	 * to provider-default auto selection. Default: true.
+	 */
+	supportsForcedToolChoice?: boolean;
 	/**
 	 * Drop reasoning fields (`reasoning_effort`, OpenRouter `reasoning`) for
 	 * the request when `tool_choice` forces a tool call. Mirrors the Anthropic

package/src/wire/gemini-headers.ts

@@ -20,6 +20,8 @@ export const ANTIGRAVITY_SYSTEM_INSTRUCTION =
 	"You are pair programming with a USER to solve their coding task. The task may require creating a new codebase, modifying or debugging an existing codebase, or simply answering a question." +
 	"**Absolute paths only**" +
 	"**Proactiveness**";
+export const ANTIGRAVITY_NO_PREAMBLE_INSTRUCTION =
+	'CRITICAL: NEVER output rule checks, formatting guidelines, constraint checklists (e.g. "No emdashes"), or your thinking/personality preambles in the final response. Output only the final response.';
 /**
  * Antigravity / Cloud Code Assist user agent. Lives in its own file so discovery
  * and usage code can read it without pulling the heavy google-gemini-cli provider