@oh-my-pi/pi-catalog 16.0.3 → 16.0.5

package/CHANGELOG.md

@@ -2,6 +2,36 @@
 
 ## [Unreleased]
 
+## [16.0.5] - 2026-06-17
+
+### Added
+
+- Added `enableGeminiThinkingLoopGuard` to OpenAI compatibility options to allow explicit opt-in or opt-out of the Gemini thinking-loop guard for OpenAI-compatible model aliases
+- Added `LITELLM_BASE_URL` as the LiteLLM provider discovery base URL fallback, with discovery caches scoped by the resolved proxy URL and explicit provider `baseUrl` config kept at higher precedence. ([#2726](https://github.com/can1357/oh-my-pi/issues/2726))
+- Added `ThinkingConfig.effortBudgets` (per-effort thinking-budget contract baked into collapsed variants) and `ANTIGRAVITY_MODEL_WIRE_PROFILES` (`maxOutputTokens` + `model_enum` per Antigravity wire id) to mirror the captured Antigravity Cloud Code Assist client request shape.
+
+### Changed
+
+- Defaulted `enableGeminiThinkingLoopGuard` from Gemini family detection for both OpenAI completions and responses compatibility specs so Gemini models now enable the thinking-loop guard automatically
+- Updated the default Gemini CLI user-agent version fallback to 0.46.0.
+- Changed the Antigravity (`google-antigravity`, daily-cloudcode-pa) gemini-3.x collapse families to the `budget` thinking transport with the client's per-tier `thinkingBudget` (3.5 Flash low/medium/high = 1000/4000/10000, 3.1 Pro low/high = 1001/10001) and corrected 3.5 Flash effort→wire routing (medium → `gemini-3.5-flash-low`, high → `gemini-3-flash-agent`). Split the shared CCA collapse table so `google-gemini-cli` (cloudcode-pa) keeps the `google-level` `thinkingLevel` transport for official Gemini CLI parity. Stale collapsed snapshots (bundled catalog, recycled `gemini-3-flash` alias) self-heal from the hand table at collapse time, and the model cache schema is bumped to v7 to invalidate pre-budget Antigravity rows.
+- Changed the Antigravity user-agent to the `antigravity/hub/<version>` format (default `2.1.4`) to match the captured client.
+
+### Fixed
+
+- Fixed `off` effort routing for `claude-opus-4-5` and `claude-opus-4-6` to use their base model IDs when thinking is disabled
+- Fixed `gemini-2.5-flash` effort routing so all non-off effort levels resolve to `gemini-2.5-flash-thinking`
+- Fixed shared variant alias provider resolution so `resolveBareVariantAlias` reports all matching providers when model aliases are present in both CCA collapse tables
+- Routed google-antigravity default baseUrl to the stable primary daily endpoint in the catalog generator and all fallback snapshots, resolving connection drops on heavy queries.
+- Fixed MiniMax M3 dialect selection so MiniMax-family OpenAI-compatible models use the MiniMax tool-call dialect instead of generic XML. ([#2759](https://github.com/can1357/oh-my-pi/issues/2759))
+- Fixed GitHub Copilot dynamic discovery to honor plan-specific API endpoints stored in structured OAuth credentials. ([#2876](https://github.com/can1357/oh-my-pi/issues/2876))
+
+## [16.0.4] - 2026-06-17
+
+### Fixed
+
+- Fixed GLM-5.2 catalog thinking metadata for Zhipu/BigModel so the top effort is exposed as `xhigh` and maps to provider-native `max`. ([#2833](https://github.com/can1357/oh-my-pi/issues/2833))
+
 ## [16.0.2] - 2026-06-16
 
 ### Fixed

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

@@ -5,6 +5,7 @@ import type { ModelSpec, OpenAICompat, ResolvedOpenAICompat, ResolvedOpenAIRespo
  */
 export declare function buildOpenAICompat(spec: ModelSpec<"openai-completions">): ResolvedOpenAICompat;
 interface OpenAIResponsesSpecLike {
+    id?: string;
     provider: string;
     name: string;
     baseUrl: string;

package/dist/types/discovery/antigravity.d.ts

@@ -1,4 +1,7 @@
 import type { ModelSpec } from "../types";
+import { type VariantCollapseTable } from "../variant-collapse";
+export declare const ANTIGRAVITY_PRIMARY_ENDPOINT = "https://daily-cloudcode-pa.googleapis.com";
+export declare const ANTIGRAVITY_SANDBOX_ENDPOINT = "https://daily-cloudcode-pa.sandbox.googleapis.com";
 /**
  * Raw model metadata returned by Antigravity's `fetchAvailableModels` endpoint.
  */
@@ -51,6 +54,12 @@ export interface FetchAntigravityDiscoveryModelsOptions {
     signal?: AbortSignal;
     /** Optional fetch implementation override for tests. */
     fetcher?: typeof fetch;
+    /**
+     * Hand collapse table to apply to the discovered list. Defaults to the
+     * Antigravity (budget-transport) table; `googleGeminiCli` passes the
+     * level-transport table so cloudcode-pa keeps `thinkingLevel`.
+     */
+    collapseTable?: VariantCollapseTable;
 }
 /**
  * Fetches discoverable Antigravity models and normalizes them into canonical model entries.

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

@@ -1,3 +1,3 @@
-export type Dialect = "glm" | "hermes" | "kimi" | "xml" | "anthropic" | "deepseek" | "harmony" | "pi" | "qwen3" | "gemini" | "gemma";
+export type Dialect = "glm" | "hermes" | "kimi" | "xml" | "anthropic" | "deepseek" | "harmony" | "pi" | "qwen3" | "gemini" | "gemma" | "minimax";
 export declare const FALLBACK_DIALECT: Dialect;
 export declare function preferredDialect(modelId: string): Dialect;

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

@@ -32,6 +32,8 @@ export declare function isMimoModelIdOrName(value: string): boolean;
  * clamp instead. Excludes M1, M3, MiniMax-Text-01, music, hailuo, voice ids.
  */
 export declare function isMinimaxM2FamilyModelId(modelId: string): boolean;
+/** MiniMax M3 family ids in bundled/default and aggregator namespace forms. */
+export declare function isMinimaxM3FamilyModelId(modelId: string): boolean;
 /**
  * OpenAI gpt-oss family (`gpt-oss-20b`, `gpt-oss-120b`, `gpt-oss:120b`,
  * `vendor/gpt-oss-…`). The Harmony reasoning format only accepts
@@ -49,6 +51,8 @@ export declare function isOpenAIModelId(modelId: string): boolean;
  * allowlist.
  */
 export declare function isReasoningGlmModelId(modelId: string): boolean;
+/** GLM-5.2+ coding SKUs accept `reasoning_effort` in addition to binary thinking. */
+export declare function isGlm52ReasoningEffortModelId(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;
 /**

package/dist/types/types.d.ts

@@ -33,6 +33,14 @@ export interface ThinkingConfig {
      * thinking is disabled. Missing keys fall back to `requestModelId ?? id`.
      */
     effortRouting?: Readonly<Partial<Record<Effort | "off", string>>>;
+    /**
+     * Per-effort thinking budget in tokens, baked at build time for collapsed
+     * variants whose upstream expects an explicit `thinkingBudget` instead of a
+     * value derived from the generic ladder (Antigravity Cloud Code Assist
+     * gemini-3.x). Request mapping prefers caller `thinkingBudgets`, then this
+     * map, then the provider default ladder. Only meaningful for `mode: "budget"`.
+     */
+    effortBudgets?: Readonly<Partial<Record<Effort, number>>>;
     /**
      * When true, a thinking-off request MUST explicitly suppress thinking on
      * the wire (google-level: `thinkingLevel: "MINIMAL"` + `includeThoughts:
@@ -137,6 +145,13 @@ export interface OpenAICompat {
     reasoningEffortMap?: Partial<Record<Effort, string>>;
     /** Whether the provider supports `stream_options: { include_usage: true }` for token usage in streaming responses. Default: true. */
     supportsUsageInStreaming?: boolean;
+    /**
+     * Enable the Gemini thinking-loop guard (pi-ai stream layer) for this model.
+     * Defaults to true when the model id classifies as the gemini family. Set
+     * explicitly to cover an opaque OpenAI-compat proxy alias (e.g. `my-model`)
+     * that routes to Gemini, or to false to opt a gemini-family id out.
+     */
+    enableGeminiThinkingLoopGuard?: boolean;
     /** Which field to use for max tokens. Default: auto-detected from URL. */
     maxTokensField?: "max_completion_tokens" | "max_tokens";
     /** Whether tool results require the `name` field. Default: auto-detected from URL. */
@@ -322,7 +337,7 @@ type ResolvedToolStrictMode = NonNullable<OpenAICompat["toolStrictMode"]> | "mix
  * `buildModel`; request handlers read fields and never detect, resolve, or
  * allocate.
  */
-export type ResolvedOpenAICompat = Required<Omit<OpenAICompat, "openRouterRouting" | "vercelGatewayRouting" | "extraBody" | "toolStrictMode" | "streamIdleTimeoutMs" | "supportsLongPromptCacheRetention" | "cacheControlFormat" | "thinkingKeep" | "strictResponsesPairing" | "requiresJuiceZeroHack" | "whenThinking">> & {
+export type ResolvedOpenAICompat = Required<Omit<OpenAICompat, "openRouterRouting" | "vercelGatewayRouting" | "extraBody" | "toolStrictMode" | "streamIdleTimeoutMs" | "supportsLongPromptCacheRetention" | "cacheControlFormat" | "thinkingKeep" | "strictResponsesPairing" | "requiresJuiceZeroHack" | "enableGeminiThinkingLoopGuard" | "whenThinking">> & {
     openRouterRouting?: OpenAICompat["openRouterRouting"];
     vercelGatewayRouting?: OpenAICompat["vercelGatewayRouting"];
     extraBody?: OpenAICompat["extraBody"];
@@ -334,6 +349,8 @@ export type ResolvedOpenAICompat = Required<Omit<OpenAICompat, "openRouterRoutin
     isOpenRouterHost: boolean;
     /** The model sits behind Vercel AI Gateway. */
     isVercelGatewayHost: boolean;
+    /** See {@link OpenAICompat.enableGeminiThinkingLoopGuard}. Set by the builder from the family classifier. */
+    enableGeminiThinkingLoopGuard?: boolean;
     /** Complete alternate view for thinking-engaged requests; swap pointers, never spread. */
     whenThinking?: ResolvedOpenAICompat;
 };
@@ -346,6 +363,8 @@ export interface ResolvedOpenAIResponsesCompat {
     strictResponsesPairing: boolean;
     requiresJuiceZeroHack: boolean;
     reasoningEffortMap: Partial<Record<Effort, string>>;
+    /** See {@link OpenAICompat.enableGeminiThinkingLoopGuard}. */
+    enableGeminiThinkingLoopGuard?: boolean;
 }
 /** Fully-resolved anthropic-messages compat view (same contract as `ResolvedOpenAICompat`). */
 export type ResolvedAnthropicCompat = Required<AnthropicCompat> & {

package/dist/types/variant-collapse.d.ts

@@ -46,12 +46,11 @@ export interface EffortVariantFamily {
 export interface VariantCollapseTable {
     families: readonly EffortVariantFamily[];
 }
-/**
- * Shared by `google-antigravity` and `google-gemini-cli` — both serve the
- * Antigravity discovery list (`fetchAntigravityDiscoveryModels`).
- */
+/** `google-antigravity` (daily-cloudcode-pa): Gemini 3.x on the budget transport. */
 export declare const ANTIGRAVITY_VARIANT_COLLAPSE_TABLE: VariantCollapseTable;
-/** Provider id → hand collapse table. Both CCA providers share one table. */
+/** `google-gemini-cli` (cloudcode-pa): Gemini 3.x on the level transport (official CLI parity). */
+export declare const GEMINI_CLI_VARIANT_COLLAPSE_TABLE: VariantCollapseTable;
+/** Provider id → hand collapse table. The CCA providers diverge on thinking transport. */
 export declare const VARIANT_COLLAPSE_TABLES: Readonly<Record<string, VariantCollapseTable>>;
 /**
  * The global automatic rule: derive an `X` + `X-thinking` family for every

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

@@ -9,7 +9,6 @@ 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
@@ -17,3 +16,19 @@ export declare const ANTIGRAVITY_NO_PREAMBLE_INSTRUCTION = "CRITICAL: NEVER outp
  * parse graph.
  */
 export declare let getAntigravityUserAgent: () => string;
+/**
+ * Per-wire-id Antigravity Cloud Code Assist request constants, captured from the
+ * real `antigravity/hub` client against `daily-cloudcode-pa`. `modelEnum` is the
+ * opaque `labels.model_enum` token the client tags each request with;
+ * `maxOutputTokens` is the fixed `generationConfig.maxOutputTokens` it sends
+ * regardless of the thinking budget. Keyed by the routed upstream wire id
+ * (post effort-routing), not the collapsed logical id. Checkpoint-only ids
+ * (e.g. `gemini-3.1-flash-lite`) are intentionally absent — this provider only
+ * emits agent requests.
+ */
+export interface AntigravityModelWireProfile {
+    modelEnum: string;
+    maxOutputTokens: number;
+}
+export declare const ANTIGRAVITY_MODEL_WIRE_PROFILES: Readonly<Record<string, AntigravityModelWireProfile>>;
+export declare function getAntigravityModelWireProfile(wireModelId: string): AntigravityModelWireProfile | undefined;

package/dist/types/wire/github-copilot.d.ts

@@ -25,9 +25,11 @@ export declare const COPILOT_API_HEADERS: {
 export type ParsedGitHubCopilotApiKey = {
     accessToken: string;
     enterpriseUrl?: string;
+    apiEndpoint?: string;
 };
 export declare function isPublicGitHubHost(host: string): boolean;
 export declare function normalizeGitHubCopilotEnterpriseDomain(input: string | undefined): string | undefined;
+export declare function normalizeGitHubCopilotApiEndpoint(input: string | undefined): string | undefined;
 export declare function parseGitHubCopilotApiKey(apiKeyRaw: string): ParsedGitHubCopilotApiKey;
 export declare function normalizeDomain(input: string): string | null;
 export declare function getGitHubCopilotBaseUrl(enterpriseDomain?: string): string;

package/package.json

@@ -1,7 +1,7 @@
 {
 	"type": "module",
 	"name": "@oh-my-pi/pi-catalog",
-	"version": "16.0.3",
+	"version": "16.0.5",
 	"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": "16.0.3",
+		"@oh-my-pi/pi-utils": "16.0.5",
 		"zod": "^4"
 	},
 	"devDependencies": {
-		"@oh-my-pi/pi-ai": "16.0.3",
+		"@oh-my-pi/pi-ai": "16.0.5",
 		"@types/bun": "^1.3.14"
 	},
 	"engines": {

package/src/compat/openai.ts

@@ -12,10 +12,12 @@ import {
 	isAnthropicNamespacedModelId,
 	isClaudeModelId,
 	isDeepseekModelIdOrName,
+	isGlm52ReasoningEffortModelId,
 	isKimiK26ModelId,
 	isKimiModelId,
 	isMimoModelIdOrName,
 	isQwenModelId,
+	modelFamilyToken,
 } from "../identity/family";
 import type { ModelSpec, OpenAICompat, ResolvedOpenAICompat, ResolvedOpenAIResponsesCompat } from "../types";
 import { applyCompatOverrides } from "./apply";
@@ -82,6 +84,7 @@ export function buildOpenAICompat(spec: ModelSpec<"openai-completions">): Resolv
 	const isCerebras = modelMatchesHost(hostModel, "cerebras");
 	const isZai = modelMatchesHost(hostModel, "zai");
 	const isZhipu = modelMatchesHost(hostModel, "zhipu");
+	const supportsZaiReasoningEffort = (isZai || isZhipu) && isGlm52ReasoningEffortModelId(spec.id);
 	const isKilo = modelMatchesHost(hostModel, "kilo");
 	const isKimiModel = isKimiModelId(spec.id);
 	const isMoonshotNative = modelMatchesHost(hostModel, "moonshotNative");
@@ -136,6 +139,8 @@ export function buildOpenAICompat(spec: ModelSpec<"openai-completions">): Resolv
 	const useMaxTokens =
 		isMistral ||
 		isMoonshotNative ||
+		isZai ||
+		isZhipu ||
 		hostMatchesUrl(baseUrl, "chutes") ||
 		hostMatchesUrl(baseUrl, "fireworks") ||
 		isDirectDeepseekApi;
@@ -202,11 +207,15 @@ export function buildOpenAICompat(spec: ModelSpec<"openai-completions">): Resolv
 		// OpenAI's reasoning-API surface.
 		supportsDeveloperRole: isOpenAIHost || isAzureHost,
 		supportsMultipleSystemMessages: supportsMultipleSystemMessagesDefault,
-		supportsReasoningEffort: !isGrok && !isZai && !isZhipu && !isXiaomiMimo,
+		supportsReasoningEffort: !isGrok && !isXiaomiMimo && (!(isZai || isZhipu) || supportsZaiReasoningEffort),
 		// GitHub Copilot's chat-completions endpoint rejects reasoning params wholesale.
 		supportsReasoningParams: provider !== "github-copilot",
 		reasoningEffortMap: {},
 		supportsUsageInStreaming: !isCerebras,
+		// pi-ai's thinking-loop guard is gemini-only; default the flag from the
+		// family classifier so OpenAI-compat proxies serving Gemini are covered.
+		// An opaque alias can opt in via `compat.enableGeminiThinkingLoopGuard`.
+		enableGeminiThinkingLoopGuard: modelFamilyToken(spec.id) === "gemini",
 		// Kimi (including via OpenRouter and Fireworks router-form IDs such as
 		// `accounts/fireworks/routers/kimi-*`) calculates TPM rate limits based on
 		// max_tokens, not actual output. The official Kimi K2 model guidance
@@ -291,6 +300,7 @@ export function buildOpenAICompat(spec: ModelSpec<"openai-completions">): Resolv
 }
 
 interface OpenAIResponsesSpecLike {
+	id?: string;
 	provider: string;
 	name: string;
 	baseUrl: string;
@@ -325,6 +335,7 @@ export function buildOpenAIResponsesCompat(spec: OpenAIResponsesSpecLike): Resol
 		strictResponsesPairing: isAzure || spec.provider === "github-copilot",
 		requiresJuiceZeroHack: spec.name.toLowerCase().startsWith("gpt-5"),
 		reasoningEffortMap: {},
+		enableGeminiThinkingLoopGuard: modelFamilyToken(spec.id ?? "") === "gemini",
 	};
 	applyCompatOverrides(compat, spec.compat);
 	return compat;

package/src/discovery/antigravity.ts

@@ -1,13 +1,16 @@
 import { z } from "zod/v4";
 import type { ModelSpec } from "../types";
 import { toPositiveNumber } from "../utils";
-import { ANTIGRAVITY_VARIANT_COLLAPSE_TABLE, collapseEffortVariants } from "../variant-collapse";
+import {
+	ANTIGRAVITY_VARIANT_COLLAPSE_TABLE,
+	collapseEffortVariants,
+	type VariantCollapseTable,
+} from "../variant-collapse";
 import { getAntigravityUserAgent } from "../wire/gemini-headers";
 
-const DEFAULT_ANTIGRAVITY_DISCOVERY_ENDPOINTS = [
-	"https://daily-cloudcode-pa.googleapis.com",
-	"https://daily-cloudcode-pa.sandbox.googleapis.com",
-] as const;
+export const ANTIGRAVITY_PRIMARY_ENDPOINT = "https://daily-cloudcode-pa.googleapis.com";
+export const ANTIGRAVITY_SANDBOX_ENDPOINT = "https://daily-cloudcode-pa.sandbox.googleapis.com";
+const DEFAULT_ANTIGRAVITY_DISCOVERY_ENDPOINTS = [ANTIGRAVITY_PRIMARY_ENDPOINT, ANTIGRAVITY_SANDBOX_ENDPOINT] as const;
 const FETCH_AVAILABLE_MODELS_PATH = "/v1internal:fetchAvailableModels";
 
 const DEFAULT_CONTEXT_WINDOW = 200_000;
@@ -157,6 +160,12 @@ export interface FetchAntigravityDiscoveryModelsOptions {
 	signal?: AbortSignal;
 	/** Optional fetch implementation override for tests. */
 	fetcher?: typeof fetch;
+	/**
+	 * Hand collapse table to apply to the discovered list. Defaults to the
+	 * Antigravity (budget-transport) table; `googleGeminiCli` passes the
+	 * level-transport table so cloudcode-pa keeps `thinkingLevel`.
+	 */
+	collapseTable?: VariantCollapseTable;
 }
 
 /**
@@ -239,7 +248,7 @@ export async function fetchAntigravityDiscoveryModels(
 		// Collapse effort-tier variants at the source so runtime discovery,
 		// the gemini-cli re-provision, and the catalog generator all see
 		// logical ids only.
-		const collapsed = collapseEffortVariants(models, ANTIGRAVITY_VARIANT_COLLAPSE_TABLE);
+		const collapsed = collapseEffortVariants(models, options.collapseTable ?? ANTIGRAVITY_VARIANT_COLLAPSE_TABLE);
 		collapsed.sort((a, b) => a.name.localeCompare(b.name) || a.id.localeCompare(b.id));
 		return collapsed;
 	}

package/src/identity/dialect.ts

@@ -11,7 +11,8 @@ export type Dialect =
 	| "pi"
 	| "qwen3"
 	| "gemini"
-	| "gemma";
+	| "gemma"
+	| "minimax";
 
 export const FALLBACK_DIALECT: Dialect = "xml";
 
@@ -31,6 +32,8 @@ export function preferredDialect(modelId: string): Dialect {
 			return "qwen3";
 		case "deepseek":
 			return "deepseek";
+		case "minimax":
+			return "minimax";
 		case "openai":
 		case "gpt-oss":
 			return "harmony";

package/src/identity/family.ts

@@ -73,6 +73,13 @@ export function isMinimaxM2FamilyModelId(modelId: string): boolean {
 	return /(?:^|[/.-])m2\d*(?:[.-]\d+)?(?:[-.:_]|$)/i.test(lower);
 }
 
+/** MiniMax M3 family ids in bundled/default and aggregator namespace forms. */
+export function isMinimaxM3FamilyModelId(modelId: string): boolean {
+	const lower = modelId.toLowerCase();
+	if (!lower.includes("minimax")) return false;
+	return /(?:^|[/._-])(?:minimax[/._-])?m3(?:[-.:_]|$)/i.test(lower);
+}
+
 /**
  * OpenAI gpt-oss family (`gpt-oss-20b`, `gpt-oss-120b`, `gpt-oss:120b`,
  * `vendor/gpt-oss-…`). The Harmony reasoning format only accepts
@@ -105,6 +112,17 @@ export function isReasoningGlmModelId(modelId: string): boolean {
 	}
 	return semverGte(glm.version, "4.5");
 }
+/** GLM-5.2+ coding SKUs accept `reasoning_effort` in addition to binary thinking. */
+export function isGlm52ReasoningEffortModelId(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, "5.2");
+}
 
 /** GLM vision SKUs — the `v` that attaches to the version (`glm-4v`, `glm-4.5v`). */
 export function isGlmVisionModelId(modelId: string): boolean {
@@ -128,7 +146,7 @@ export function modelFamilyToken(modelId: string): string {
 	if (isOpenAIModelId(modelId)) return "openai";
 	if (isKimiModelId(modelId)) return "kimi";
 	if (isQwenModelId(modelId)) return "qwen";
-	if (isMinimaxM2FamilyModelId(modelId)) return "minimax";
+	if (isMinimaxM2FamilyModelId(modelId) || isMinimaxM3FamilyModelId(modelId)) return "minimax";
 	if (isOpenAIGptOssModelId(modelId)) return "gpt-oss";
 	if (isDeepseekModelIdOrName(modelId)) return "deepseek";
 	if (isMimoModelIdOrName(modelId)) return "mimo";

package/src/model-cache.ts

@@ -7,12 +7,14 @@ import { getModelDbPath } from "@oh-my-pi/pi-utils";
 import type { Api, Model, ModelSpec } from "./types";
 
 // Rows persist ModelSpec JSON (sparse `compat`, never the resolved record);
-// the model manager rebuilds via `buildModel` on load. v6 invalidates rows
-// that may contain the retired unknown-limit sentinels (222222/8888); v5
-// invalidated rows predating effort-tier variant collapsing (raw
-// `-low`/`-high`/`-thinking` member ids); v4 dropped the pre-efforts
-// ThinkingConfig shape.
-const CACHE_SCHEMA_VERSION = 6;
+// the model manager rebuilds via `buildModel` on load. v7 invalidates rows
+// predating the Antigravity Gemini budget-mode migration (cached specs still
+// carrying `thinking.mode: "google-level"` and the old 3.5-flash effort
+// routing); v6 invalidates rows that may contain the retired unknown-limit
+// sentinels (222222/8888); v5 invalidated rows predating effort-tier variant
+// collapsing (raw `-low`/`-high`/`-thinking` member ids); v4 dropped the
+// pre-efforts ThinkingConfig shape.
+const CACHE_SCHEMA_VERSION = 7;
 
 interface CacheRow {
 	provider_id: string;

package/src/model-thinking.ts

@@ -23,6 +23,7 @@ import {
 import {
 	findThinkingVariantToken,
 	isDeepseekModelIdOrName,
+	isGlm52ReasoningEffortModelId,
 	isMinimaxM2FamilyModelId,
 	isOpenAIGptOssModelId,
 	supportsAdaptiveThinkingDisplay,
@@ -76,6 +77,13 @@ const DEEPSEEK_REASONING_EFFORT_MAP: Readonly<EffortMap> = {
 const FIREWORKS_REASONING_EFFORT_MAP: Readonly<EffortMap> = {
 	[Effort.Minimal]: "none",
 };
+const ZAI_GLM_52_REASONING_EFFORT_MAP: Readonly<EffortMap> = {
+	[Effort.Minimal]: "none",
+	[Effort.Low]: "high",
+	[Effort.Medium]: "high",
+	[Effort.High]: "high",
+	[Effort.XHigh]: "max",
+};
 
 /**
  * Effort → wire-value map for the 5-tier adaptive scale (Opus 4.7+ and
@@ -259,11 +267,19 @@ function sameEffortList(left: readonly Effort[], right: readonly Effort[]): bool
 }
 
 function getModelDefinedEfforts<TApi extends Api>(spec: ModelSpec<TApi>): readonly Effort[] | undefined {
+	if (spec.api === "openai-completions" && isZaiGlm52ReasoningEffortModel(spec)) {
+		return DEFAULT_REASONING_EFFORTS_WITH_XHIGH;
+	}
 	return spec.api === "openai-completions" && (isMinimaxM2FamilyModelId(spec.id) || isOpenAIGptOssModelId(spec.id))
 		? LOW_MEDIUM_HIGH_REASONING_EFFORTS
 		: undefined;
 }
 
+function isZaiGlm52ReasoningEffortModel<TApi extends Api>(spec: ModelSpec<TApi>): boolean {
+	if (!isGlm52ReasoningEffortModelId(spec.id)) return false;
+	return modelMatchesHost(spec, "zai") || modelMatchesHost(spec, "zhipu");
+}
+
 function readCompatEffortMap(compat: CompatOf<Api>): EffortMap | undefined {
 	if (compat === undefined || !("reasoningEffortMap" in compat)) {
 		return undefined;
@@ -288,6 +304,9 @@ function inferDetectedEffortMap<TApi extends Api>(
 	if (spec.provider === "groq" && spec.id === "qwen/qwen3-32b") {
 		return GROQ_QWEN3_32B_REASONING_EFFORT_MAP;
 	}
+	if (isZaiGlm52ReasoningEffortModel(spec)) {
+		return ZAI_GLM_52_REASONING_EFFORT_MAP;
+	}
 	if (isDeepseekReasoningModel(spec)) {
 		return DEEPSEEK_REASONING_EFFORT_MAP;
 	}