@oh-my-pi/pi-catalog 16.0.6 → 16.0.8

package/CHANGELOG.md

@@ -2,6 +2,22 @@
 
 ## [Unreleased]
 
+## [16.0.8] - 2026-06-18
+
+### Changed
+
+- Refactored model family ID predicates and capability checkers to use a shared, uniform process-lifetime `memo` utility to eliminate caching boilerplate.
+
+### Fixed
+
+- Fixed LM Studio dynamic discovery to use native `/api/v0/models` metadata so VLM models advertise image input. ([#2945](https://github.com/can1357/oh-my-pi/issues/2945))
+
+## [16.0.7] - 2026-06-18
+
+### Fixed
+
+- Fixed MiniMax Anthropic-compatible M2/M3 thinking metadata to expose the adaptive transport and keep M2 mandatory reasoning floored ([#2928](https://github.com/can1357/oh-my-pi/issues/2928)).
+
 ## [16.0.6] - 2026-06-18
 
 ### Added

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

@@ -41,7 +41,6 @@ export interface UnknownModel {
     id: string;
 }
 export type ParsedModel = GeminiModel | AnthropicModel | OpenAIModel | UnknownModel;
-/** 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 const parseGeminiModel: (modelId: string) => GeminiModel | null;

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

@@ -7,27 +7,27 @@
  * here.
  */
 /** Kimi family ids in any namespace form (`moonshotai/kimi-*`, `kimi-k2.6`, `vendor/kimi.x`). */
-export declare function isKimiModelId(modelId: string): boolean;
+export declare const isKimiModelId: (modelId: string) => boolean;
 /** Kimi K2.6 specifically, including router ids that spell the version `k2p6`. */
-export declare function isKimiK26ModelId(modelId: string): boolean;
+export declare const isKimiK26ModelId: (modelId: string) => boolean;
 /** Claude ids in any namespace form (`claude-*`, `vendor/claude.x`). */
-export declare function isClaudeModelId(modelId: string): boolean;
+export declare const isClaudeModelId: (modelId: string) => boolean;
 /** `anthropic/`-namespaced ids (aggregator catalogs like OpenRouter). */
-export declare function isAnthropicNamespacedModelId(modelId: string): boolean;
+export declare const isAnthropicNamespacedModelId: (modelId: string) => boolean;
 /** Qwen family ids (substring match — Qwen SKUs have no stable prefix shape). */
-export declare function isQwenModelId(modelId: string): boolean;
+export declare const isQwenModelId: (modelId: string) => boolean;
 /** Gemma open-weights family (`gemma-3-27b-it`, `google/gemma-4-E2B-it`, `gemma2-9b`). */
-export declare function isGemmaModelId(modelId: string): boolean;
+export declare const isGemmaModelId: (modelId: string) => boolean;
 /** DeepSeek family by id or display name (proxies often rename the id but keep the name). */
-export declare function isDeepseekModelIdOrName(value: string): boolean;
+export declare const isDeepseekModelIdOrName: (modelId: string) => boolean;
 /** Xiaomi MiMo family by id or display name. */
-export declare function isMimoModelIdOrName(value: string): boolean;
+export declare const isMimoModelIdOrName: (modelId: string) => boolean;
 /**
  * Grok SKUs that expose the wire `reasoning.effort` dial. Other Grok reasoners
  * (e.g. `grok-build`, `grok-4.20-0309-reasoning`) think natively but reject the
  * param, so callers must omit reasoning effort for them.
  */
-export declare function isGrokReasoningEffortCapable(modelId: string): boolean;
+export declare const isGrokReasoningEffortCapable: (modelId: string) => boolean;
 /**
  * MiniMax M2-generation family (M2, M2.1, M2.5, M2.7, including `-highspeed`/
  * `-lightning`/`-her`/`-turbo` variants, dotless aliases like `minimax-m21`,
@@ -37,18 +37,18 @@ export declare function isGrokReasoningEffortCapable(modelId: string): boolean;
  * `minimal` to `none` (Fireworks) or expects the full 5-tier scale must
  * clamp instead. Excludes M1, M3, MiniMax-Text-01, music, hailuo, voice ids.
  */
-export declare function isMinimaxM2FamilyModelId(modelId: string): boolean;
+export declare const isMinimaxM2FamilyModelId: (modelId: string) => boolean;
 /** MiniMax M3 family ids in bundled/default and aggregator namespace forms. */
-export declare function isMinimaxM3FamilyModelId(modelId: string): boolean;
+export declare const 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
  * `low|medium|high` for `reasoning_effort` and rejects `minimal`, `xhigh`,
  * and `none`.
  */
-export declare function isOpenAIGptOssModelId(modelId: string): boolean;
+export declare const isOpenAIGptOssModelId: (modelId: string) => boolean;
 /** OpenAI model ids (gpt-*, o1-*, o3-*, o4-*, or prefixed with openai/). */
-export declare function isOpenAIModelId(modelId: string): boolean;
+export declare const isOpenAIModelId: (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
@@ -56,11 +56,11 @@ export declare function isOpenAIModelId(modelId: string): boolean;
  * keeps newly-bumped integers (`glm-5.3`, `glm-6`, …) covered without a per-id
  * allowlist.
  */
-export declare function isReasoningGlmModelId(modelId: string): boolean;
+export declare const isReasoningGlmModelId: (modelId: string) => boolean;
 /** GLM-5.2+ coding SKUs accept `reasoning_effort` in addition to binary thinking. */
-export declare function isGlm52ReasoningEffortModelId(modelId: string): boolean;
+export declare const 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;
+export declare const 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,
@@ -72,7 +72,7 @@ export declare function isGlmVisionModelId(modelId: string): boolean;
  * 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;
+export declare const 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
@@ -80,13 +80,13 @@ export declare function modelFamilyToken(modelId: string): string;
  * dashed version forms both match while bare dated ids
  * (`claude-opus-4-20250514` = Opus 4.0) stay excluded.
  */
-export declare function supportsAdaptiveThinkingDisplay(modelId: string): boolean;
+export declare const supportsAdaptiveThinkingDisplay: (modelId: string) => boolean;
 /**
  * Returns true for Anthropic models with Opus 4.7+/Fable/Mythos API restrictions:
  * - Sampling parameters (temperature/top_p/top_k) return 400 error
  * - Thinking content is omitted by default (needs display: "summarized")
  */
-export declare function hasOpus47ApiRestrictions(modelId: string): boolean;
+export declare const hasOpus47ApiRestrictions: (modelId: string) => boolean;
 /**
  * Mid-conversation `role: "system"` messages (system instructions appended at
  * non-first positions in the `messages` array) are supported starting with
@@ -94,8 +94,8 @@ export declare function hasOpus47ApiRestrictions(modelId: string): boolean;
  * models reject the role.
  * @see https://platform.claude.com/docs/en/build-with-claude/mid-conversation-system-messages
  */
-export declare function supportsMidConversationSystemMessages(modelId: string): boolean;
-export declare function isAnthropicFableOrMythosModel(modelId: string): boolean;
+export declare const supportsMidConversationSystemMessages: (modelId: string) => boolean;
+export declare const isAnthropicFableOrMythosModel: (modelId: string) => boolean;
 /** Thinking-variant token location inside a model id. */
 export interface ThinkingVariantToken {
     index: number;
@@ -115,4 +115,4 @@ export declare function findThinkingVariantToken(modelId: string): ThinkingVaria
  * token exists or nothing would remain. Callers MUST verify the result names
  * a live model.
  */
-export declare function stripThinkingVariantToken(modelId: string): string | undefined;
+export declare const stripThinkingVariantToken: (modelId: string) => string | undefined;

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

@@ -63,7 +63,7 @@ export declare function mapEffortToGoogleThinkingLevel(effort: Effort): "MINIMAL
  * Maps a normalized thinking effort to Anthropic adaptive effort values via
  * the model's baked `thinking.effortMap` (identity for unmapped efforts).
  */
-export declare function mapEffortToAnthropicAdaptiveEffort<TApi extends Api>(model: ApiModel<TApi>, effort: Effort): "low" | "medium" | "high" | "xhigh" | "max";
+export declare function mapEffortToAnthropicAdaptiveEffort<TApi extends Api>(model: ApiModel<TApi>, effort: Effort): "low" | "medium" | "high" | "xhigh" | "max" | "adaptive";
 /**
  * Resolves the upstream wire model id for a request at the given effort
  * (`undefined` = thinking off). Collapsed effort-tier variants route through

package/dist/types/provider-models/openai-compat.d.ts

@@ -265,6 +265,18 @@ export interface KimiCodeModelManagerConfig {
     fetch?: FetchImpl;
 }
 export declare function kimiCodeModelManagerOptions(config?: KimiCodeModelManagerConfig): ModelManagerOptions<"openai-completions">;
+/** Native LM Studio metadata keyed by model id from `/api/v0/models`. */
+export interface LmStudioNativeModelMetadata {
+    input: ("text" | "image")[];
+    contextWindow?: number;
+}
+/** Options for LM Studio's optional native metadata probe. */
+export interface LmStudioNativeModelMetadataOptions {
+    headers?: Record<string, string>;
+    signal?: AbortSignal;
+}
+/** Fetches LM Studio native model metadata used to mark VLM models as image-capable. */
+export declare function fetchLmStudioNativeModelMetadata(baseUrl: string, fetchImpl?: FetchImpl, options?: LmStudioNativeModelMetadataOptions): Promise<Map<string, LmStudioNativeModelMetadata> | null>;
 export interface LmStudioModelManagerConfig {
     apiKey?: string;
     baseUrl?: string;

package/package.json

@@ -1,7 +1,7 @@
 {
 	"type": "module",
 	"name": "@oh-my-pi/pi-catalog",
-	"version": "16.0.6",
+	"version": "16.0.8",
 	"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": "16.0.6",
+		"@oh-my-pi/pi-utils": "16.0.8",
 		"arktype": "^2.2.0",
 		"zod": "^4"
 	},
 	"devDependencies": {
-		"@oh-my-pi/pi-ai": "16.0.6",
+		"@oh-my-pi/pi-ai": "16.0.8",
 		"@types/bun": "^1.3.14"
 	},
 	"engines": {

package/src/hosts.ts

@@ -16,7 +16,7 @@ interface HostClassSpec {
 	readonly providers?: readonly string[];
 	/** Provider-id prefixes that imply this host class (e.g. `xiaomi-token-plan-`). */
 	readonly providerPrefixes?: readonly string[];
-	/** Case-insensitive substrings matched against the base URL. */
+	/** Lowercase ASCII substrings matched case-insensitively against the base URL. */
 	readonly urlMarkers: readonly string[];
 	// Strict hostname matching is intentionally not modeled here: the one
 	// auth-sensitive consumer (Anthropic official-endpoint) parses the URL
@@ -68,9 +68,8 @@ export type KnownHost = keyof typeof KNOWN_HOSTS;
 export function hostMatchesUrl(baseUrl: string | undefined, host: KnownHost): boolean {
 	if (!baseUrl) return false;
 	const spec: HostClassSpec = KNOWN_HOSTS[host];
-	const normalized = baseUrl.toLowerCase();
 	for (const marker of spec.urlMarkers) {
-		if (normalized.includes(marker)) return true;
+		if (includesAsciiCaseInsensitive(baseUrl, marker)) return true;
 	}
 	return false;
 }
@@ -91,6 +90,19 @@ export function modelMatchesHost(model: { provider: string; baseUrl: string }, h
 	return hostMatchesUrl(model.baseUrl, host);
 }
 
+function includesAsciiCaseInsensitive(value: string, lowerNeedle: string): boolean {
+	const needleLength = lowerNeedle.length;
+	const end = value.length - needleLength;
+	for (let start = 0; start <= end; start++) {
+		let offset = 0;
+		for (; offset < needleLength; offset++) {
+			if ((value.charCodeAt(start + offset) | 0x20) !== lowerNeedle.charCodeAt(offset)) break;
+		}
+		if (offset === needleLength) return true;
+	}
+	return false;
+}
+
 // --- Endpoint-shape predicates (URL path/verb shapes, not vendor hosts) ---
 
 /** Vertex AI express-mode OpenAI-compatible endpoint (`…/endpoints/openapi`). */

package/src/identity/classify.ts

@@ -51,9 +51,15 @@ export interface UnknownModel {
 export type ParsedModel = GeminiModel | AnthropicModel | OpenAIModel | UnknownModel;
 
 /** Strip a provider namespace prefix (`openai/gpt-5.4` → `gpt-5.4`). */
+// Cache keyed by model id (a bounded set of bundled/aggregator ids), so no eviction is needed.
+const bareModelIdCache = new Map<string, string>();
 export function bareModelId(modelId: string): string {
+	const cached = bareModelIdCache.get(modelId);
+	if (cached !== undefined) return cached;
 	const p = modelId.lastIndexOf("/");
-	return p !== -1 ? modelId.slice(p + 1) : modelId;
+	const result = p !== -1 ? modelId.slice(p + 1) : modelId;
+	bareModelIdCache.set(modelId, result);
+	return result;
 }
 
 export function parseKnownModel(modelId: string): ParsedModel {

package/src/identity/family.ts

@@ -16,45 +16,58 @@ import {
 	semverGte,
 } from "./classify";
 
+/** Bounded process-lifetime cache memo helper. */
+function memo<T>(fn: (modelId: string) => T): (modelId: string) => T {
+	const cache = new Map<string, T>();
+	return (modelId: string) => {
+		if (cache.has(modelId)) {
+			return cache.get(modelId) as T;
+		}
+		const result = fn(modelId);
+		cache.set(modelId, result);
+		return result;
+	};
+}
+
 /** Kimi family ids in any namespace form (`moonshotai/kimi-*`, `kimi-k2.6`, `vendor/kimi.x`). */
-export function isKimiModelId(modelId: string): boolean {
+export const isKimiModelId = memo((modelId: string): boolean => {
 	return modelId.includes("moonshotai/kimi") || /(^|\/)kimi[-.]/i.test(modelId);
-}
+});
 
 /** Kimi K2.6 specifically, including router ids that spell the version `k2p6`. */
-export function isKimiK26ModelId(modelId: string): boolean {
+export const isKimiK26ModelId = memo((modelId: string): boolean => {
 	return /(^|\/)kimi-k2(?:\.6|p6)(?:[-:]|$)/i.test(modelId);
-}
+});
 
 /** Claude ids in any namespace form (`claude-*`, `vendor/claude.x`). */
-export function isClaudeModelId(modelId: string): boolean {
+export const isClaudeModelId = memo((modelId: string): boolean => {
 	return /(^|\/)claude[-.]/i.test(modelId);
-}
+});
 
 /** `anthropic/`-namespaced ids (aggregator catalogs like OpenRouter). */
-export function isAnthropicNamespacedModelId(modelId: string): boolean {
+export const isAnthropicNamespacedModelId = memo((modelId: string): boolean => {
 	return /(^|\/)anthropic\//i.test(modelId);
-}
+});
 
 /** Qwen family ids (substring match — Qwen SKUs have no stable prefix shape). */
-export function isQwenModelId(modelId: string): boolean {
+export const isQwenModelId = memo((modelId: string): boolean => {
 	return modelId.toLowerCase().includes("qwen");
-}
+});
 
 /** Gemma open-weights family (`gemma-3-27b-it`, `google/gemma-4-E2B-it`, `gemma2-9b`). */
-export function isGemmaModelId(modelId: string): boolean {
+export const isGemmaModelId = memo((modelId: string): boolean => {
 	return /(^|\/)gemma[-.]?\d/i.test(modelId);
-}
+});
 
 /** DeepSeek family by id or display name (proxies often rename the id but keep the name). */
-export function isDeepseekModelIdOrName(value: string): boolean {
+export const isDeepseekModelIdOrName = memo((value: string): boolean => {
 	return value.toLowerCase().includes("deepseek");
-}
+});
 
 /** Xiaomi MiMo family by id or display name. */
-export function isMimoModelIdOrName(value: string): boolean {
+export const isMimoModelIdOrName = memo((value: string): boolean => {
 	return value.toLowerCase().includes("mimo");
-}
+});
 
 const GROK_EFFORT_CAPABLE_PREFIXES = ["grok-3-mini", "grok-4.20-multi-agent", "grok-4.3"] as const;
 
@@ -63,11 +76,11 @@ const GROK_EFFORT_CAPABLE_PREFIXES = ["grok-3-mini", "grok-4.20-multi-agent", "g
  * (e.g. `grok-build`, `grok-4.20-0309-reasoning`) think natively but reject the
  * param, so callers must omit reasoning effort for them.
  */
-export function isGrokReasoningEffortCapable(modelId: string): boolean {
+export const isGrokReasoningEffortCapable = memo((modelId: string): boolean => {
 	const bare = bareModelId(modelId).trim().toLowerCase();
 	if (!bare) return false;
 	return GROK_EFFORT_CAPABLE_PREFIXES.some(prefix => bare.startsWith(prefix));
-}
+});
 
 /**
  * MiniMax M2-generation family (M2, M2.1, M2.5, M2.7, including `-highspeed`/
@@ -78,20 +91,20 @@ export function isGrokReasoningEffortCapable(modelId: string): boolean {
  * `minimal` to `none` (Fireworks) or expects the full 5-tier scale must
  * clamp instead. Excludes M1, M3, MiniMax-Text-01, music, hailuo, voice ids.
  */
-export function isMinimaxM2FamilyModelId(modelId: string): boolean {
+export const isMinimaxM2FamilyModelId = memo((modelId: string): boolean => {
 	const lower = modelId.toLowerCase();
 	if (!lower.includes("minimax")) return false;
 	// Boundary-delimited `m2` token followed by zero or more digits (dotless
 	// variants like `m21`/`m25`/`m27`) and an optional dotted minor version.
 	return /(?:^|[/.-])m2\d*(?:[.-]\d+)?(?:[-.:_]|$)/i.test(lower);
-}
+});
 
 /** MiniMax M3 family ids in bundled/default and aggregator namespace forms. */
-export function isMinimaxM3FamilyModelId(modelId: string): boolean {
+export const isMinimaxM3FamilyModelId = memo((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`,
@@ -99,14 +112,14 @@ export function isMinimaxM3FamilyModelId(modelId: string): boolean {
  * `low|medium|high` for `reasoning_effort` and rejects `minimal`, `xhigh`,
  * and `none`.
  */
-export function isOpenAIGptOssModelId(modelId: string): boolean {
+export const isOpenAIGptOssModelId = memo((modelId: string): boolean => {
 	return /(^|\/)gpt-oss[-:]/i.test(modelId);
-}
+});
 
 /** OpenAI model ids (gpt-*, o1-*, o3-*, o4-*, or prefixed with openai/). */
-export function isOpenAIModelId(modelId: string): boolean {
+export const isOpenAIModelId = memo((modelId: string): boolean => {
 	return /(^|\/)(gpt|o1|o3|o4)[-.]/i.test(modelId) || modelId.toLowerCase().includes("openai/");
-}
+});
 
 /**
  * Reasoning-capable GLM coding SKUs: glm-4.5 and up on the base / `-air` /
@@ -115,7 +128,7 @@ export function isOpenAIModelId(modelId: string): boolean {
  * keeps newly-bumped integers (`glm-5.3`, `glm-6`, …) covered without a per-id
  * allowlist.
  */
-export function isReasoningGlmModelId(modelId: string): boolean {
+export const isReasoningGlmModelId = memo((modelId: string): boolean => {
 	const glm = parseGlmModel(bareModelId(modelId));
 	if (!glm || glm.vision) {
 		return false;
@@ -124,9 +137,10 @@ export function isReasoningGlmModelId(modelId: string): boolean {
 		return false;
 	}
 	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 {
+export const isGlm52ReasoningEffortModelId = memo((modelId: string): boolean => {
 	const glm = parseGlmModel(bareModelId(modelId));
 	if (!glm || glm.vision) {
 		return false;
@@ -135,12 +149,13 @@ export function isGlm52ReasoningEffortModelId(modelId: string): boolean {
 		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 {
+export const isGlmVisionModelId = memo((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,
@@ -152,7 +167,7 @@ export function isGlmVisionModelId(modelId: string): boolean {
  * 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 {
+export const modelFamilyToken = memo((modelId: string): string => {
 	const parsed = parseKnownModel(modelId);
 	if (parsed.family !== "unknown") return parsed.family;
 	if (isClaudeModelId(modelId) || isAnthropicNamespacedModelId(modelId)) return "anthropic";
@@ -166,7 +181,7 @@ export function modelFamilyToken(modelId: string): string {
 	if (isGemmaModelId(modelId)) return "gemma";
 	if (parseGlmModel(bareModelId(modelId))) return "glm";
 	return "";
-}
+});
 
 /**
  * Adaptive thinking `display` is supported starting with Claude Opus 4.7 and
@@ -175,23 +190,23 @@ export function modelFamilyToken(modelId: string): string {
  * dashed version forms both match while bare dated ids
  * (`claude-opus-4-20250514` = Opus 4.0) stay excluded.
  */
-export function supportsAdaptiveThinkingDisplay(modelId: string): boolean {
+export const supportsAdaptiveThinkingDisplay = memo((modelId: string): boolean => {
 	const parsed = parseAnthropicModel(bareModelId(modelId));
 	if (!parsed) return false;
 	if (isFableOrMythos(parsed.kind)) return semverGte(parsed.version, "5");
 	return parsed.kind === "opus" && semverGte(parsed.version, "4.7");
-}
+});
 
 /**
  * Returns true for Anthropic models with Opus 4.7+/Fable/Mythos API restrictions:
  * - Sampling parameters (temperature/top_p/top_k) return 400 error
  * - Thinking content is omitted by default (needs display: "summarized")
  */
-export function hasOpus47ApiRestrictions(modelId: string): boolean {
+export const hasOpus47ApiRestrictions = memo((modelId: string): boolean => {
 	const parsed = parseAnthropicModel(bareModelId(modelId));
 	if (!parsed) return false;
 	return (parsed.kind === "opus" && semverGte(parsed.version, "4.7")) || isFableOrMythos(parsed.kind);
-}
+});
 
 /**
  * Mid-conversation `role: "system"` messages (system instructions appended at
@@ -200,16 +215,16 @@ export function hasOpus47ApiRestrictions(modelId: string): boolean {
  * models reject the role.
  * @see https://platform.claude.com/docs/en/build-with-claude/mid-conversation-system-messages
  */
-export function supportsMidConversationSystemMessages(modelId: string): boolean {
+export const supportsMidConversationSystemMessages = memo((modelId: string): boolean => {
 	const parsed = parseAnthropicModel(bareModelId(modelId));
 	if (!parsed) return false;
 	return (parsed.kind === "opus" && semverGte(parsed.version, "4.8")) || isFableOrMythos(parsed.kind);
-}
+});
 
-export function isAnthropicFableOrMythosModel(modelId: string): boolean {
+export const isAnthropicFableOrMythosModel = memo((modelId: string): boolean => {
 	const parsed = parseAnthropicModel(bareModelId(modelId));
 	return parsed !== null && isFableOrMythos(parsed.kind);
-}
+});
 
 /** Thinking-variant token location inside a model id. */
 export interface ThinkingVariantToken {
@@ -245,9 +260,9 @@ export function findThinkingVariantToken(modelId: string): ThinkingVariantToken
  * token exists or nothing would remain. Callers MUST verify the result names
  * a live model.
  */
-export function stripThinkingVariantToken(modelId: string): string | undefined {
+export const stripThinkingVariantToken = memo((modelId: string): string | undefined => {
 	const token = findThinkingVariantToken(modelId);
 	if (!token) return undefined;
 	const stripped = modelId.slice(0, token.index) + modelId.slice(token.index + token.length);
 	return stripped.length > 0 ? stripped : undefined;
-}
+});

package/src/model-thinking.ts

@@ -25,6 +25,7 @@ import {
 	isDeepseekModelIdOrName,
 	isGlm52ReasoningEffortModelId,
 	isMinimaxM2FamilyModelId,
+	isMinimaxM3FamilyModelId,
 	isOpenAIGptOssModelId,
 	supportsAdaptiveThinkingDisplay,
 } from "./identity/family";
@@ -113,6 +114,12 @@ export const ANTHROPIC_ADAPTIVE_EFFORT_MAP_4_TIER: Readonly<Partial<Record<Effor
 	[Effort.XHigh]: "max",
 };
 
+const MINIMAX_ANTHROPIC_ADAPTIVE_EFFORT_MAP: Readonly<EffortMap> = {
+	[Effort.Low]: "adaptive",
+	[Effort.Medium]: "adaptive",
+	[Effort.High]: "adaptive",
+};
+
 // ---------------------------------------------------------------------------
 // Build-time derivation (buildModel + catalog generator only)
 // ---------------------------------------------------------------------------
@@ -295,6 +302,10 @@ function isOllamaCloudGlm52ReasoningEffortModel<TApi extends Api>(spec: ModelSpe
 	return spec.api === "ollama-chat" && spec.provider === "ollama-cloud" && isGlm52ReasoningEffortModelId(spec.id);
 }
 
+function isMinimaxReasoningModelOnAnthropicEndpoint<TApi extends Api>(spec: ModelSpec<TApi>): boolean {
+	return spec.api === "anthropic-messages" && (isMinimaxM2FamilyModelId(spec.id) || isMinimaxM3FamilyModelId(spec.id));
+}
+
 function readCompatEffortMap(compat: CompatOf<Api>): EffortMap | undefined {
 	if (compat === undefined || !("reasoningEffortMap" in compat)) {
 		return undefined;
@@ -309,6 +320,9 @@ function inferDetectedEffortMap<TApi extends Api>(
 	mode: ThinkingConfig["mode"],
 ): EffortMap | undefined {
 	if (mode === "anthropic-adaptive") {
+		if (isMinimaxReasoningModelOnAnthropicEndpoint(spec)) {
+			return MINIMAX_ANTHROPIC_ADAPTIVE_EFFORT_MAP;
+		}
 		return anthropicModelHasRealXHighEffort(spec, parsedModel)
 			? ANTHROPIC_ADAPTIVE_EFFORT_MAP_5_TIER
 			: ANTHROPIC_ADAPTIVE_EFFORT_MAP_4_TIER;
@@ -446,6 +460,9 @@ function inferAnthropicSupportedEfforts<TApi extends Api>(
 }
 
 function inferFallbackEfforts<TApi extends Api>(spec: ModelSpec<TApi>, compat: CompatOf<TApi>): readonly Effort[] {
+	if (isMinimaxReasoningModelOnAnthropicEndpoint(spec)) {
+		return LOW_MEDIUM_HIGH_REASONING_EFFORTS;
+	}
 	if (spec.api === "anthropic-messages") {
 		return DEFAULT_REASONING_EFFORTS_WITH_XHIGH;
 	}
@@ -488,6 +505,9 @@ function inferThinkingControlMode<TApi extends Api>(
 				: "budget";
 
 		case "anthropic-messages":
+			if (isMinimaxReasoningModelOnAnthropicEndpoint(spec)) {
+				return "anthropic-adaptive";
+			}
 			if (parsedModel.family === "anthropic") {
 				if (semverGte(parsedModel.version, "4.6")) {
 					return "anthropic-adaptive";
@@ -626,9 +646,15 @@ export function mapEffortToGoogleThinkingLevel(effort: Effort): "MINIMAL" | "LOW
 export function mapEffortToAnthropicAdaptiveEffort<TApi extends Api>(
 	model: ApiModel<TApi>,
 	effort: Effort,
-): "low" | "medium" | "high" | "xhigh" | "max" {
+): "low" | "medium" | "high" | "xhigh" | "max" | "adaptive" {
 	const supported = requireSupportedEffort(model, effort);
-	return (model.thinking?.effortMap?.[supported] ?? supported) as "low" | "medium" | "high" | "xhigh" | "max";
+	return (model.thinking?.effortMap?.[supported] ?? supported) as
+		| "low"
+		| "medium"
+		| "high"
+		| "xhigh"
+		| "max"
+		| "adaptive";
 }
 
 /**