@oh-my-pi/pi-catalog 16.1.7 → 16.1.9

package/CHANGELOG.md

@@ -2,6 +2,24 @@
 
 ## [Unreleased]
 
+## [16.1.9] - 2026-06-21
+
+### Fixed
+
+- Fixed the `moonshot` provider with no path to the Kimi China API: model discovery now honors a `MOONSHOT_BASE_URL` override (redirecting to `api.moonshot.cn`), and `KIMI_API_KEY` resolves as a fallback for `MOONSHOT_API_KEY`. ([#2883](https://github.com/can1357/oh-my-pi/issues/2883))
+- Fixed LiteLLM model discovery preserving colliding models.dev transport metadata (for example `ollama-cloud` `deepseek-v4-flash`) instead of keeping the LiteLLM `openai-completions` provider transport. ([#3162](https://github.com/can1357/oh-my-pi/issues/3162))
+
+### Removed
+
+- Removed bundled Wafer Pass (`wafer-pass`) catalog entries and generation support; Wafer Serverless remains available as `wafer-serverless`.
+
+## [16.1.8] - 2026-06-20
+
+### Fixed
+
+- Fixed Fireworks-hosted Qwen turns (e.g. `fireworks/qwen3.7-plus`) failing with `400 Extra inputs are not permitted, field: 'enable_thinking'`. Fireworks serves Qwen3 with controllable thinking via OpenAI-style `reasoning_effort` and rejects the top-level `enable_thinking` boolean that Alibaba DashScope speaks; `buildOpenAICompat` was selecting `thinkingFormat: "qwen"` from the `qwen` id pattern regardless of host. Fireworks-hosted Qwen models now resolve to `thinkingFormat: "openai"`.
+- Fixed MiMo models on OpenAI-compatible gateways to expose only accepted `low`, `medium`, and `high` reasoning tiers and map unsupported raw `minimal`/`xhigh` requests to safe wire values. ([#2864](https://github.com/can1357/oh-my-pi/issues/2864))
+
 ## [16.1.7] - 2026-06-20
 
 ### Fixed

package/dist/types/fireworks-model-id.d.ts

@@ -8,3 +8,16 @@ export declare function toFireworksWireModelId(modelId: string): string;
  */
 export declare function toFirepassPublicModelId(modelId: string): string;
 export declare function toFirepassWireModelId(modelId: string): string;
+/**
+ * Public-id suffix marking a Fireworks "Fast" serving-path variant. Fast is a
+ * higher-throughput route (100+ tok/s) exposed under a dedicated router id
+ * (`accounts/fireworks/routers/<id>-fast`), not a separate model — same weights,
+ * higher price, no Priority tier. We keep a friendly `<id>-fast` public id and
+ * translate it to the router wire form at request time (compat
+ * `wireModelIdMode: "firepass"`). See https://docs.fireworks.ai/serverless/serving-paths.
+ */
+export declare const FIREWORKS_FAST_SUFFIX = "-fast";
+/** True for a Fireworks public model id that selects the Fast serving path. */
+export declare function isFireworksFastModelId(modelId: string): boolean;
+/** Strip the Fast suffix to recover the base (Standard-tier) model id. */
+export declare function toFireworksBaseModelId(modelId: string): string;

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

@@ -173,7 +173,7 @@ export declare const CATALOG_PROVIDERS: readonly [{
 }, {
     readonly id: "moonshot";
     readonly defaultModel: "kimi-k2.7-code";
-    readonly envVars: readonly ["MOONSHOT_API_KEY"];
+    readonly envVars: readonly ["MOONSHOT_API_KEY", "KIMI_API_KEY"];
     readonly createModelManagerOptions: (config: ModelManagerConfig) => import("..").ModelManagerOptions<"openai-completions", unknown>;
     readonly catalogDiscovery: {
         readonly label: "Moonshot";
@@ -310,15 +310,6 @@ export declare const CATALOG_PROVIDERS: readonly [{
         readonly label: "vLLM";
         readonly allowUnauthenticated: true;
     };
-}, {
-    readonly id: "wafer-pass";
-    readonly defaultModel: "GLM-5.1";
-    readonly envVars: readonly ["WAFER_PASS_API_KEY"];
-    readonly createModelManagerOptions: (config: ModelManagerConfig) => import("..").ModelManagerOptions<"openai-completions", unknown>;
-    readonly catalogDiscovery: {
-        readonly label: "Wafer Pass";
-        readonly oauthProvider: "wafer-pass";
-    };
 }, {
     readonly id: "wafer-serverless";
     readonly defaultModel: "GLM-5.1";

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

@@ -179,6 +179,13 @@ export declare const KIMI_K27_CODE_RECOMMENDED_MAX_TOKENS = 32768;
 export declare function isKimiK27CodeModelId(modelId: string): boolean;
 export declare function clampKimiK27CodeMaxTokens(modelId: string, candidate: number): number;
 export declare function clampKimiK27CodeMaxTokens(modelId: string, candidate: number | null): number | null;
+/**
+ * Build the Fireworks Fast seed by projecting each base bundled spec into a
+ * `<id>-fast` variant. Pushed into the generated catalog (Fast routers never
+ * appear in the serverless control-plane list, so discovery cannot surface
+ * them) and deduped behind any identical previous-snapshot entry.
+ */
+export declare function buildFireworksFastSeed(): ModelSpec<"openai-completions">[];
 /**
  * Fireworks DeepSeek V4 accepts effort via `reasoning_effort` but rejects the
  * DeepSeek-native binary `thinking` toggle when both are present.
@@ -208,7 +215,6 @@ export interface WaferModelManagerConfig {
     baseUrl?: string;
     fetch?: FetchImpl;
 }
-export declare function waferPassModelManagerOptions(config?: WaferModelManagerConfig): ModelManagerOptions<"openai-completions">;
 export declare function waferServerlessModelManagerOptions(config?: WaferModelManagerConfig): ModelManagerOptions<"openai-completions">;
 export interface MistralModelManagerConfig {
     apiKey?: string;

package/package.json

@@ -1,7 +1,7 @@
 {
 	"type": "module",
 	"name": "@oh-my-pi/pi-catalog",
-	"version": "16.1.7",
+	"version": "16.1.9",
 	"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": "16.1.7",
+		"@oh-my-pi/pi-utils": "16.1.9",
 		"arktype": "^2.2.0",
 		"zod": "^4"
 	},
 	"devDependencies": {
-		"@oh-my-pi/pi-ai": "16.1.7",
+		"@oh-my-pi/pi-ai": "16.1.9",
 		"@types/bun": "^1.3.14"
 	},
 	"engines": {

package/src/compat/openai.ts

@@ -7,6 +7,7 @@
  * complete alternate views. Request handlers read `model.compat` fields and
  * never detect, resolve, or allocate.
  */
+import { isFireworksFastModelId } from "../fireworks-model-id";
 import { hostMatchesUrl, modelMatchesHost } from "../hosts";
 import {
 	isAnthropicNamespacedModelId,
@@ -130,6 +131,16 @@ const OPENCODE_WHEN_THINKING: NonNullable<OpenAICompat["whenThinking"]> = {
 	reasoningContentField: "reasoning_content",
 };
 
+const MIMO_REASONING_EFFORT_MAP: NonNullable<OpenAICompat["reasoningEffortMap"]> = {
+	minimal: "low",
+	xhigh: "high",
+};
+
+function mergeMimoReasoningEffortMap(compat: ResolvedOpenAISharedCompat, enabled: boolean): void {
+	if (!enabled) return;
+	compat.reasoningEffortMap = { ...MIMO_REASONING_EFFORT_MAP, ...compat.reasoningEffortMap };
+}
+
 function detectStrictModeSupport(provider: string, baseUrl: string): boolean {
 	if (
 		provider === "openai" ||
@@ -184,6 +195,8 @@ export function buildOpenAICompat(spec: ModelSpec<"openai-completions">): Resolv
 	const lowerName = (spec.name ?? "").toLowerCase();
 	const isXiaomiHost = modelMatchesHost(hostModel, "xiaomi");
 	const isXiaomiMimo = isXiaomiHost && (isMimoModelIdOrName(spec.id) || isMimoModelIdOrName(spec.name ?? ""));
+	const isMimoReasoningEffortModel =
+		!isXiaomiHost && (isMimoModelIdOrName(spec.id) || isMimoModelIdOrName(spec.name ?? ""));
 	// OpenCode Zen's `big-pickle` is a DeepSeek reasoning alias; the upstream
 	// 400s come from DeepSeek and require exact reasoning_content replay.
 	const isOpenCodeDeepseekAlias =
@@ -238,17 +251,21 @@ export function buildOpenAICompat(spec: ModelSpec<"openai-completions">): Resolv
 	const isGroqHost = modelMatchesHost(hostModel, "groq");
 	const isCopilotHost = provider === "github-copilot";
 	const isZenmuxHost = provider === "zenmux";
-	// Endpoints that MUST receive a single system block. MiniMax's OpenAI
-	// endpoint returns error 2013 on multiple system messages; Alibaba's
-	// Dashscope and Qwen Portal serve Qwen models whose chat template
-	// raises "System message must be at the beginning" if any system
-	// message appears past index 0.
+	// Endpoints/models that MUST receive a single system block. MiniMax's OpenAI
+	// endpoint returns error 2013 on multiple system messages; the Qwen 3.5+ chat
+	// template raises "System message must be at the beginning" / 500s with an
+	// internal_server_error when any system block appears past index 0. That
+	// template ships with the weights, so every Qwen-serving vLLM/SGLang host
+	// hits it — confirmed on Alibaba Dashscope, Qwen Portal, and Fireworks
+	// (`fireworks/qwen3.7-plus` 500'd on two leading system blocks). Gate on the
+	// Qwen family itself, not per-host: coalescing only trades away KV-cache reuse.
 	const isMiniMaxHost = modelMatchesHost(hostModel, "minimax");
 	const isQwenPortal = modelMatchesHost(hostModel, "qwenPortal");
 	const supportsMultipleSystemMessagesDefault =
 		!isMiniMaxHost &&
 		!isAlibaba &&
 		!isQwenPortal &&
+		!isQwen &&
 		(isOpenAIHost ||
 			isAzureHost ||
 			isOpenRouter ||
@@ -276,8 +293,12 @@ export function buildOpenAICompat(spec: ModelSpec<"openai-completions">): Resolv
 					? DEEPSEEK_REASONING_STREAM_IDLE_TIMEOUT_MS
 					: undefined;
 
+	// Fireworks "Fast" variants (`<id>-fast`) are served from the router
+	// namespace (`accounts/fireworks/routers/<id>-fast`), like Fire Pass, rather
+	// than the `models/` namespace the rest of the `fireworks` provider uses.
+	const isFireworksFastRouter = provider === "fireworks" && isFireworksFastModelId(spec.id);
 	const wireModelIdMode: ResolvedOpenAISharedCompat["wireModelIdMode"] =
-		provider === "firepass"
+		provider === "firepass" || isFireworksFastRouter
 			? "firepass"
 			: provider === "fireworks"
 				? "fireworks"
@@ -291,9 +312,11 @@ export function buildOpenAICompat(spec: ModelSpec<"openai-completions">): Resolv
 				? "openrouter"
 				: isQwen && isNvidiaNim
 					? "qwen-chat-template"
-					: isAlibaba || isQwen
-						? "qwen"
-						: "openai";
+					: isQwen && isFireworks
+						? "openai"
+						: isAlibaba || isQwen
+							? "qwen"
+							: "openai";
 
 	const compat: ResolvedOpenAICompat = {
 		supportsStore: !isNonStandard,
@@ -308,7 +331,7 @@ export function buildOpenAICompat(spec: ModelSpec<"openai-completions">): Resolv
 		supportsReasoningEffort: !isGrok && !isXiaomiMimo && (!(isZai || isZhipu) || supportsZaiReasoningEffort),
 		// GitHub Copilot's chat-completions endpoint rejects reasoning params wholesale.
 		supportsReasoningParams: provider !== "github-copilot",
-		reasoningEffortMap: {},
+		reasoningEffortMap: isMimoReasoningEffortModel ? MIMO_REASONING_EFFORT_MAP : {},
 		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.
@@ -400,6 +423,7 @@ export function buildOpenAICompat(spec: ModelSpec<"openai-completions">): Resolv
 		compat.omitReasoningEffort = true;
 	}
 	mergeOllamaReasoningEffortMap(compat, provider, spec.reasoning);
+	mergeMimoReasoningEffortMap(compat, isMimoReasoningEffortModel);
 
 	const whenThinkingPolicy =
 		spec.compat?.whenThinking ?? (isOpenCodeProvider && spec.reasoning ? OPENCODE_WHEN_THINKING : undefined);
@@ -413,6 +437,7 @@ export function buildOpenAICompat(spec: ModelSpec<"openai-completions">): Resolv
 			variant.omitReasoningEffort = true;
 		}
 		mergeOllamaReasoningEffortMap(variant, provider, spec.reasoning);
+		mergeMimoReasoningEffortMap(variant, isMimoReasoningEffortModel);
 		compat.whenThinking = variant;
 	}
 

package/src/fireworks-model-id.ts

@@ -28,3 +28,23 @@ export function toFirepassWireModelId(modelId: string): string {
 	const stripped = modelId.startsWith(FIREPASS_WIRE_PREFIX) ? modelId.slice(FIREPASS_WIRE_PREFIX.length) : modelId;
 	return `${FIREPASS_WIRE_PREFIX}${stripped.replace(VERSION_DOT_PATTERN, "p")}`;
 }
+
+/**
+ * Public-id suffix marking a Fireworks "Fast" serving-path variant. Fast is a
+ * higher-throughput route (100+ tok/s) exposed under a dedicated router id
+ * (`accounts/fireworks/routers/<id>-fast`), not a separate model — same weights,
+ * higher price, no Priority tier. We keep a friendly `<id>-fast` public id and
+ * translate it to the router wire form at request time (compat
+ * `wireModelIdMode: "firepass"`). See https://docs.fireworks.ai/serverless/serving-paths.
+ */
+export const FIREWORKS_FAST_SUFFIX = "-fast";
+
+/** True for a Fireworks public model id that selects the Fast serving path. */
+export function isFireworksFastModelId(modelId: string): boolean {
+	return modelId.endsWith(FIREWORKS_FAST_SUFFIX);
+}
+
+/** Strip the Fast suffix to recover the base (Standard-tier) model id. */
+export function toFireworksBaseModelId(modelId: string): string {
+	return modelId.endsWith(FIREWORKS_FAST_SUFFIX) ? modelId.slice(0, -FIREWORKS_FAST_SUFFIX.length) : modelId;
+}

package/src/model-thinking.ts

@@ -24,6 +24,7 @@ import {
 	findThinkingVariantToken,
 	isDeepseekModelIdOrName,
 	isGlm52ReasoningEffortModelId,
+	isMimoModelIdOrName,
 	isMinimaxM2FamilyModelId,
 	isMinimaxM3FamilyModelId,
 	isOpenAIGptOssModelId,
@@ -89,6 +90,10 @@ const ZAI_GLM_52_REASONING_EFFORT_MAP: Readonly<EffortMap> = {
 const GLM_52_XHIGH_MAX_EFFORT_MAP: Readonly<EffortMap> = {
 	[Effort.XHigh]: "max",
 };
+const MIMO_REASONING_EFFORT_MAP: Readonly<EffortMap> = {
+	[Effort.Minimal]: "low",
+	[Effort.XHigh]: "high",
+};
 
 /**
  * Effort → wire-value map for the 5-tier adaptive scale (Opus 4.7+ and
@@ -296,7 +301,10 @@ function getModelDefinedEfforts<TApi extends Api>(
 			return GLM_52_HIGH_MAX_REASONING_EFFORTS;
 		}
 	}
-	return isOpenAICompatReasoningApi(spec.api) && (isMinimaxM2FamilyModelId(spec.id) || isOpenAIGptOssModelId(spec.id))
+	return isOpenAICompatReasoningApi(spec.api) &&
+		(isMinimaxM2FamilyModelId(spec.id) ||
+			isOpenAIGptOssModelId(spec.id) ||
+			isOpenAICompatMimoReasoningEffortModel(spec, compat))
 		? LOW_MEDIUM_HIGH_REASONING_EFFORTS
 		: undefined;
 }
@@ -309,6 +317,19 @@ function isMinimaxReasoningModelOnAnthropicEndpoint<TApi extends Api>(spec: Mode
 	return spec.api === "anthropic-messages" && (isMinimaxM2FamilyModelId(spec.id) || isMinimaxM3FamilyModelId(spec.id));
 }
 
+function isOpenAICompatMimoReasoningEffortModel<TApi extends Api>(
+	spec: ModelSpec<TApi>,
+	compat: CompatOf<TApi>,
+): boolean {
+	if (!isOpenAICompatReasoningApi(spec.api)) return false;
+	if (!isMimoModelIdOrName(spec.id) && !isMimoModelIdOrName(spec.name ?? "")) return false;
+	const resolved = compat as ResolvedOpenAICompat | undefined;
+	return (
+		(resolved?.thinkingFormat === "openai" || resolved?.thinkingFormat === "openrouter") &&
+		resolved.supportsReasoningEffort
+	);
+}
+
 function readCompatEffortMap(compat: CompatOf<Api>): EffortMap | undefined {
 	if (compat === undefined || !("reasoningEffortMap" in compat)) {
 		return undefined;
@@ -364,6 +385,8 @@ function inferDetectedEffortMap<TApi extends Api>(
 		map = GROQ_QWEN3_32B_REASONING_EFFORT_MAP;
 	} else if (isDeepseekReasoningModel(spec)) {
 		map = DEEPSEEK_REASONING_EFFORT_MAP;
+	} else if (isOpenAICompatMimoReasoningEffortModel(spec, compat)) {
+		map = MIMO_REASONING_EFFORT_MAP;
 	} else if (modelMatchesHost(spec, "openrouter")) {
 		map = getOpenRouterAnthropicReasoningEffortMap(spec.id);
 	} else if (modelMatchesHost(spec, "fireworks")) {
@@ -485,6 +508,8 @@ function inferAnthropicSupportedEfforts<TApi extends Api>(
 }
 
 function inferFallbackEfforts<TApi extends Api>(spec: ModelSpec<TApi>, compat: CompatOf<TApi>): readonly Effort[] {
+	const modelDefinedEfforts = getModelDefinedEfforts(spec, compat);
+	if (modelDefinedEfforts !== undefined) return modelDefinedEfforts;
 	if (isMinimaxReasoningModelOnAnthropicEndpoint(spec)) {
 		return LOW_MEDIUM_HIGH_REASONING_EFFORTS;
 	}

package/src/models.json

@@ -7208,11 +7208,9 @@
 			"thinking": {
 				"mode": "effort",
 				"efforts": [
-					"minimal",
 					"low",
 					"medium",
-					"high",
-					"xhigh"
+					"high"
 				]
 			}
 		},
@@ -7238,11 +7236,9 @@
 			"thinking": {
 				"mode": "effort",
 				"efforts": [
-					"minimal",
 					"low",
 					"medium",
-					"high",
-					"xhigh"
+					"high"
 				]
 			}
 		},
@@ -7267,11 +7263,9 @@
 			"thinking": {
 				"mode": "effort",
 				"efforts": [
-					"minimal",
 					"low",
 					"medium",
-					"high",
-					"xhigh"
+					"high"
 				]
 			}
 		}
@@ -14801,6 +14795,38 @@
 				}
 			}
 		},
+		"glm-5.1-fast": {
+			"id": "glm-5.1-fast",
+			"name": "GLM-5.1 Fast",
+			"api": "openai-completions",
+			"provider": "fireworks",
+			"baseUrl": "https://api.fireworks.ai/inference/v1",
+			"reasoning": true,
+			"input": [
+				"text"
+			],
+			"cost": {
+				"input": 2.8,
+				"output": 8.8,
+				"cacheRead": 0.52,
+				"cacheWrite": 0
+			},
+			"contextWindow": 202752,
+			"maxTokens": 131072,
+			"thinking": {
+				"mode": "effort",
+				"efforts": [
+					"minimal",
+					"low",
+					"medium",
+					"high",
+					"xhigh"
+				],
+				"effortMap": {
+					"minimal": "none"
+				}
+			}
+		},
 		"glm-5.2": {
 			"id": "glm-5.2",
 			"name": "GLM-5.2",
@@ -14947,6 +14973,39 @@
 				}
 			}
 		},
+		"kimi-k2.6-fast": {
+			"id": "kimi-k2.6-fast",
+			"name": "Kimi K2.6 Fast",
+			"api": "openai-completions",
+			"provider": "fireworks",
+			"baseUrl": "https://api.fireworks.ai/inference/v1",
+			"reasoning": true,
+			"input": [
+				"text",
+				"image"
+			],
+			"cost": {
+				"input": 2,
+				"output": 8,
+				"cacheRead": 0.3,
+				"cacheWrite": 0
+			},
+			"contextWindow": 262144,
+			"maxTokens": 32768,
+			"thinking": {
+				"mode": "effort",
+				"efforts": [
+					"minimal",
+					"low",
+					"medium",
+					"high",
+					"xhigh"
+				],
+				"effortMap": {
+					"minimal": "none"
+				}
+			}
+		},
 		"kimi-k2.7-code": {
 			"id": "kimi-k2.7-code",
 			"name": "Kimi K2.7 Code",
@@ -14980,6 +15039,39 @@
 				}
 			}
 		},
+		"kimi-k2.7-code-fast": {
+			"id": "kimi-k2.7-code-fast",
+			"name": "Kimi K2.7 Code Fast",
+			"api": "openai-completions",
+			"provider": "fireworks",
+			"baseUrl": "https://api.fireworks.ai/inference/v1",
+			"reasoning": true,
+			"input": [
+				"text",
+				"image"
+			],
+			"cost": {
+				"input": 1.9,
+				"output": 8,
+				"cacheRead": 0.38,
+				"cacheWrite": 0
+			},
+			"contextWindow": 262144,
+			"maxTokens": 32768,
+			"thinking": {
+				"mode": "effort",
+				"efforts": [
+					"minimal",
+					"low",
+					"medium",
+					"high",
+					"xhigh"
+				],
+				"effortMap": {
+					"minimal": "none"
+				}
+			}
+		},
 		"minimax-m2.5": {
 			"id": "minimax-m2.5",
 			"name": "MiniMax M2.5",
@@ -20276,11 +20368,9 @@
 			"thinking": {
 				"mode": "effort",
 				"efforts": [
-					"minimal",
 					"low",
 					"medium",
-					"high",
-					"xhigh"
+					"high"
 				]
 			}
 		},
@@ -30625,11 +30715,9 @@
 			"thinking": {
 				"mode": "effort",
 				"efforts": [
-					"minimal",
 					"low",
 					"medium",
-					"high",
-					"xhigh"
+					"high"
 				]
 			}
 		},
@@ -30655,11 +30743,9 @@
 			"thinking": {
 				"mode": "effort",
 				"efforts": [
-					"minimal",
 					"low",
 					"medium",
-					"high",
-					"xhigh"
+					"high"
 				]
 			}
 		},
@@ -30703,11 +30789,9 @@
 			"thinking": {
 				"mode": "effort",
 				"efforts": [
-					"minimal",
 					"low",
 					"medium",
-					"high",
-					"xhigh"
+					"high"
 				]
 			}
 		},
@@ -30752,11 +30836,9 @@
 			"thinking": {
 				"mode": "effort",
 				"efforts": [
-					"minimal",
 					"low",
 					"medium",
-					"high",
-					"xhigh"
+					"high"
 				]
 			}
 		},
@@ -30781,11 +30863,9 @@
 			"thinking": {
 				"mode": "effort",
 				"efforts": [
-					"minimal",
 					"low",
 					"medium",
-					"high",
-					"xhigh"
+					"high"
 				]
 			}
 		},
@@ -31172,7 +31252,7 @@
 	"kimi-code": {
 		"kimi-for-coding": {
 			"id": "kimi-for-coding",
-			"name": "Kimi For Coding",
+			"name": "K2.7 Code",
 			"api": "openai-completions",
 			"provider": "kimi-code",
 			"baseUrl": "https://api.kimi.com/coding/v1",
@@ -49119,11 +49199,9 @@
 			"thinking": {
 				"mode": "effort",
 				"efforts": [
-					"minimal",
 					"low",
 					"medium",
-					"high",
-					"xhigh"
+					"high"
 				],
 				"effortRouting": {
 					"off": "xiaomi/mimo-v2-flash",
@@ -49183,11 +49261,9 @@
 			"thinking": {
 				"mode": "effort",
 				"efforts": [
-					"minimal",
 					"low",
 					"medium",
-					"high",
-					"xhigh"
+					"high"
 				],
 				"effortRouting": {
 					"off": "xiaomi/mimo-v2-flash-original",
@@ -49248,11 +49324,9 @@
 			"thinking": {
 				"mode": "effort",
 				"efforts": [
-					"minimal",
 					"low",
 					"medium",
-					"high",
-					"xhigh"
+					"high"
 				]
 			}
 		},
@@ -49277,11 +49351,9 @@
 			"thinking": {
 				"mode": "effort",
 				"efforts": [
-					"minimal",
 					"low",
 					"medium",
-					"high",
-					"xhigh"
+					"high"
 				]
 			}
 		},
@@ -49307,11 +49379,9 @@
 			"thinking": {
 				"mode": "effort",
 				"efforts": [
-					"minimal",
 					"low",
 					"medium",
-					"high",
-					"xhigh"
+					"high"
 				]
 			}
 		},
@@ -49336,11 +49406,9 @@
 			"thinking": {
 				"mode": "effort",
 				"efforts": [
-					"minimal",
 					"low",
 					"medium",
-					"high",
-					"xhigh"
+					"high"
 				]
 			}
 		},
@@ -56842,11 +56910,9 @@
 			"thinking": {
 				"mode": "effort",
 				"efforts": [
-					"minimal",
 					"low",
 					"medium",
-					"high",
-					"xhigh"
+					"high"
 				]
 			},
 			"compat": {
@@ -56874,11 +56940,9 @@
 			"thinking": {
 				"mode": "effort",
 				"efforts": [
-					"minimal",
 					"low",
 					"medium",
-					"high",
-					"xhigh"
+					"high"
 				]
 			},
 			"compat": {
@@ -56910,11 +56974,9 @@
 			"thinking": {
 				"mode": "effort",
 				"efforts": [
-					"minimal",
 					"low",
 					"medium",
-					"high",
-					"xhigh"
+					"high"
 				]
 			}
 		},
@@ -56942,11 +57004,9 @@
 			"thinking": {
 				"mode": "effort",
 				"efforts": [
-					"minimal",
 					"low",
 					"medium",
-					"high",
-					"xhigh"
+					"high"
 				]
 			}
 		},
@@ -58575,11 +58635,9 @@
 			"thinking": {
 				"mode": "effort",
 				"efforts": [
-					"minimal",
 					"low",
 					"medium",
-					"high",
-					"xhigh"
+					"high"
 				]
 			}
 		},
@@ -58605,11 +58663,9 @@
 			"thinking": {
 				"mode": "effort",
 				"efforts": [
-					"minimal",
 					"low",
 					"medium",
-					"high",
-					"xhigh"
+					"high"
 				]
 			}
 		},
@@ -58634,11 +58690,9 @@
 			"thinking": {
 				"mode": "effort",
 				"efforts": [
-					"minimal",
 					"low",
 					"medium",
-					"high",
-					"xhigh"
+					"high"
 				]
 			}
 		},
@@ -58664,11 +58718,9 @@
 			"thinking": {
 				"mode": "effort",
 				"efforts": [
-					"minimal",
 					"low",
 					"medium",
-					"high",
-					"xhigh"
+					"high"
 				]
 			}
 		},
@@ -67414,7 +67466,6 @@
 			"thinking": {
 				"mode": "effort",
 				"efforts": [
-					"minimal",
 					"low",
 					"medium",
 					"high"
@@ -67443,7 +67494,6 @@
 			"thinking": {
 				"mode": "effort",
 				"efforts": [
-					"minimal",
 					"low",
 					"medium",
 					"high"
@@ -67471,7 +67521,6 @@
 			"thinking": {
 				"mode": "effort",
 				"efforts": [
-					"minimal",
 					"low",
 					"medium",
 					"high"
@@ -67500,7 +67549,6 @@
 			"thinking": {
 				"mode": "effort",
 				"efforts": [
-					"minimal",
 					"low",
 					"medium",
 					"high"
@@ -67528,7 +67576,6 @@
 			"thinking": {
 				"mode": "effort",
 				"efforts": [
-					"minimal",
 					"low",
 					"medium",
 					"high"
@@ -72124,11 +72171,9 @@
 			"thinking": {
 				"mode": "effort",
 				"efforts": [
-					"minimal",
 					"low",
 					"medium",
-					"high",
-					"xhigh"
+					"high"
 				]
 			}
 		},
@@ -77545,64 +77590,6 @@
 			}
 		}
 	},
-	"wafer-pass": {
-		"GLM-5.1": {
-			"id": "GLM-5.1",
-			"name": "GLM-5.1",
-			"api": "openai-completions",
-			"provider": "wafer-pass",
-			"baseUrl": "https://pass.wafer.ai/v1",
-			"reasoning": true,
-			"input": [
-				"text"
-			],
-			"cost": {
-				"input": 0,
-				"output": 0,
-				"cacheRead": 0,
-				"cacheWrite": 0
-			},
-			"contextWindow": 202752,
-			"maxTokens": 65536,
-			"compat": {
-				"supportsDeveloperRole": false,
-				"thinkingFormat": "zai",
-				"reasoningContentField": "reasoning_content"
-			},
-			"thinking": {
-				"mode": "effort",
-				"efforts": [
-					"minimal",
-					"low",
-					"medium",
-					"high"
-				]
-			}
-		},
-		"Qwen3.5-397B-A17B": {
-			"id": "Qwen3.5-397B-A17B",
-			"name": "Qwen3.5-397B-A17B",
-			"api": "openai-completions",
-			"provider": "wafer-pass",
-			"baseUrl": "https://pass.wafer.ai/v1",
-			"reasoning": false,
-			"input": [
-				"text",
-				"image"
-			],
-			"cost": {
-				"input": 0,
-				"output": 0,
-				"cacheRead": 0,
-				"cacheWrite": 0
-			},
-			"contextWindow": 262144,
-			"maxTokens": 65536,
-			"compat": {
-				"supportsDeveloperRole": false
-			}
-		}
-	},
 	"wafer-serverless": {
 		"deepseek-v4-flash": {
 			"id": "deepseek-v4-flash",
@@ -84061,11 +84048,9 @@
 			"thinking": {
 				"mode": "effort",
 				"efforts": [
-					"minimal",
 					"low",
 					"medium",
-					"high",
-					"xhigh"
+					"high"
 				]
 			}
 		},
@@ -84090,11 +84075,9 @@
 			"thinking": {
 				"mode": "effort",
 				"efforts": [
-					"minimal",
 					"low",
 					"medium",
-					"high",
-					"xhigh"
+					"high"
 				]
 			}
 		},
@@ -84120,11 +84103,9 @@
 			"thinking": {
 				"mode": "effort",
 				"efforts": [
-					"minimal",
 					"low",
 					"medium",
-					"high",
-					"xhigh"
+					"high"
 				]
 			}
 		},
@@ -84149,11 +84130,9 @@
 			"thinking": {
 				"mode": "effort",
 				"efforts": [
-					"minimal",
 					"low",
 					"medium",
-					"high",
-					"xhigh"
+					"high"
 				]
 			}
 		},
@@ -84179,11 +84158,9 @@
 			"thinking": {
 				"mode": "effort",
 				"efforts": [
-					"minimal",
 					"low",
 					"medium",
-					"high",
-					"xhigh"
+					"high"
 				]
 			}
 		},
@@ -84208,11 +84185,9 @@
 			"thinking": {
 				"mode": "effort",
 				"efforts": [
-					"minimal",
 					"low",
 					"medium",
-					"high",
-					"xhigh"
+					"high"
 				]
 			}
 		},
@@ -84637,6 +84612,39 @@
 		}
 	},
 	"zhipu-coding-plan": {
+		"glm-4.5": {
+			"id": "glm-4.5",
+			"name": "glm-4.5",
+			"api": "openai-completions",
+			"provider": "zhipu-coding-plan",
+			"baseUrl": "https://open.bigmodel.cn/api/coding/paas/v4",
+			"reasoning": true,
+			"input": [
+				"text"
+			],
+			"cost": {
+				"input": 0,
+				"output": 0,
+				"cacheRead": 0,
+				"cacheWrite": 0
+			},
+			"contextWindow": 131072,
+			"maxTokens": 98304,
+			"thinking": {
+				"mode": "effort",
+				"efforts": [
+					"minimal",
+					"low",
+					"medium",
+					"high"
+				]
+			},
+			"compat": {
+				"thinkingFormat": "zai",
+				"reasoningContentField": "reasoning_content",
+				"supportsDeveloperRole": false
+			}
+		},
 		"glm-4.5-air": {
 			"id": "glm-4.5-air",
 			"name": "GLM-4.5-Air",
@@ -84670,6 +84678,39 @@
 				]
 			}
 		},
+		"glm-4.6": {
+			"id": "glm-4.6",
+			"name": "glm-4.6",
+			"api": "openai-completions",
+			"provider": "zhipu-coding-plan",
+			"baseUrl": "https://open.bigmodel.cn/api/coding/paas/v4",
+			"reasoning": true,
+			"input": [
+				"text"
+			],
+			"cost": {
+				"input": 0,
+				"output": 0,
+				"cacheRead": 0,
+				"cacheWrite": 0
+			},
+			"contextWindow": 202752,
+			"maxTokens": 131072,
+			"thinking": {
+				"mode": "effort",
+				"efforts": [
+					"minimal",
+					"low",
+					"medium",
+					"high"
+				]
+			},
+			"compat": {
+				"thinkingFormat": "zai",
+				"reasoningContentField": "reasoning_content",
+				"supportsDeveloperRole": false
+			}
+		},
 		"glm-4.6v": {
 			"id": "glm-4.6v",
 			"name": "GLM-4.6V",
@@ -84737,6 +84778,39 @@
 				]
 			}
 		},
+		"glm-5": {
+			"id": "glm-5",
+			"name": "GLM-5",
+			"api": "openai-completions",
+			"provider": "zhipu-coding-plan",
+			"baseUrl": "https://open.bigmodel.cn/api/coding/paas/v4",
+			"reasoning": true,
+			"input": [
+				"text"
+			],
+			"cost": {
+				"input": 0,
+				"output": 0,
+				"cacheRead": 0,
+				"cacheWrite": 0
+			},
+			"contextWindow": 204800,
+			"maxTokens": 131072,
+			"thinking": {
+				"mode": "effort",
+				"efforts": [
+					"minimal",
+					"low",
+					"medium",
+					"high"
+				]
+			},
+			"compat": {
+				"thinkingFormat": "zai",
+				"reasoningContentField": "reasoning_content",
+				"supportsDeveloperRole": false
+			}
+		},
 		"glm-5-turbo": {
 			"id": "glm-5-turbo",
 			"name": "GLM-5-Turbo",

package/src/provider-models/descriptors.ts

@@ -41,7 +41,6 @@ import {
 	veniceModelManagerOptions,
 	vercelAiGatewayModelManagerOptions,
 	vllmModelManagerOptions,
-	waferPassModelManagerOptions,
 	waferServerlessModelManagerOptions,
 	xaiModelManagerOptions,
 	xaiOAuthModelManagerOptions,
@@ -219,7 +218,9 @@ export const CATALOG_PROVIDERS = [
 	{
 		id: "moonshot",
 		defaultModel: "kimi-k2.7-code",
-		envVars: ["MOONSHOT_API_KEY"],
+		// KIMI_API_KEY is the most intuitive name for a Kimi/Moonshot key; accept it
+		// as a fallback so China users need not learn MOONSHOT_API_KEY. (#2883)
+		envVars: ["MOONSHOT_API_KEY", "KIMI_API_KEY"],
 		createModelManagerOptions: (config: ModelManagerConfig) => moonshotModelManagerOptions(config),
 		catalogDiscovery: { label: "Moonshot" },
 	},
@@ -347,13 +348,6 @@ export const CATALOG_PROVIDERS = [
 		createModelManagerOptions: (config: ModelManagerConfig) => vllmModelManagerOptions(config),
 		catalogDiscovery: { label: "vLLM", allowUnauthenticated: true },
 	},
-	{
-		id: "wafer-pass",
-		defaultModel: "GLM-5.1",
-		envVars: ["WAFER_PASS_API_KEY"],
-		createModelManagerOptions: (config: ModelManagerConfig) => waferPassModelManagerOptions(config),
-		catalogDiscovery: { label: "Wafer Pass", oauthProvider: "wafer-pass" },
-	},
 	{
 		id: "wafer-serverless",
 		defaultModel: "GLM-5.1",

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

@@ -4,7 +4,7 @@ import {
 	type OpenAICompatibleModelRecord,
 } from "../discovery/openai-compatible";
 import { Effort } from "../effort";
-import { toFireworksPublicModelId } from "../fireworks-model-id";
+import { FIREWORKS_FAST_SUFFIX, toFireworksPublicModelId } from "../fireworks-model-id";
 import { isGlmVisionModelId, isGrokReasoningEffortCapable, isReasoningGlmModelId } from "../identity/family";
 import type { ModelManagerOptions } from "../model-manager";
 import { getBundledModels } from "../models";
@@ -197,6 +197,8 @@ function mapWithBundledReference<TApi extends Api>(
 		...reference,
 		id: defaults.id,
 		name,
+		api: defaults.api,
+		provider: defaults.provider,
 		baseUrl: defaults.baseUrl,
 		contextWindow: toPositiveNumber(entry.context_length, reference.contextWindow),
 		maxTokens: toPositiveNumber(entry.max_completion_tokens, reference.maxTokens),
@@ -1258,6 +1260,51 @@ export function clampKimiK27CodeMaxTokens(modelId: string, candidate: number | n
 	return isKimiK27CodeModelId(modelId) ? Math.min(candidate, KIMI_K27_CODE_RECOMMENDED_MAX_TOKENS) : candidate;
 }
 
+/**
+ * Fireworks Fast variants we surface. Each inherits the base model's
+ * limits/modalities/thinking and overrides only the cost with the Standard-column
+ * Fast prices from the Serverless pricing table; `cacheWrite` stays 0 (Fireworks
+ * bills no cache-write). Derived from the bundled base entries so metadata stays
+ * in lockstep, and the runtime auto-falls back to the base id on a failed fast
+ * request. See https://docs.fireworks.ai/serverless/pricing.
+ */
+const FIREWORKS_FAST_VARIANT_SPECS: ReadonlyArray<{
+	base: string;
+	name: string;
+	cost: { input: number; output: number; cacheRead: number };
+}> = [
+	{ base: "kimi-k2.7-code", name: "Kimi K2.7 Code Fast", cost: { input: 1.9, output: 8, cacheRead: 0.38 } },
+	{ base: "kimi-k2.6", name: "Kimi K2.6 Fast", cost: { input: 2, output: 8, cacheRead: 0.3 } },
+	{ base: "glm-5.1", name: "GLM-5.1 Fast", cost: { input: 2.8, output: 8.8, cacheRead: 0.52 } },
+];
+
+/**
+ * Build the Fireworks Fast seed by projecting each base bundled spec into a
+ * `<id>-fast` variant. Pushed into the generated catalog (Fast routers never
+ * appear in the serverless control-plane list, so discovery cannot surface
+ * them) and deduped behind any identical previous-snapshot entry.
+ */
+export function buildFireworksFastSeed(): ModelSpec<"openai-completions">[] {
+	const bundled = createBundledReferenceMap<"openai-completions">("fireworks");
+	const seeds: ModelSpec<"openai-completions">[] = [];
+	for (const variant of FIREWORKS_FAST_VARIANT_SPECS) {
+		const base = bundled.get(variant.base);
+		if (!base) continue;
+		seeds.push({
+			...base,
+			id: `${variant.base}${FIREWORKS_FAST_SUFFIX}`,
+			name: variant.name,
+			cost: {
+				input: variant.cost.input,
+				output: variant.cost.output,
+				cacheRead: variant.cost.cacheRead,
+				cacheWrite: 0,
+			},
+		});
+	}
+	return seeds;
+}
+
 /**
  * Fireworks DeepSeek V4 accepts effort via `reasoning_effort` but rejects the
  * DeepSeek-native binary `thinking` toggle when both are present.
@@ -1523,7 +1570,7 @@ export function firepassModelManagerOptions(
 }
 
 // ---------------------------------------------------------------------------
-// 7.7 Wafer (Pass + Serverless)
+// 7.7 Wafer Serverless
 // ---------------------------------------------------------------------------
 
 export interface WaferModelManagerConfig {
@@ -1536,13 +1583,14 @@ const WAFER_DEFAULT_BASE_URL = "https://pass.wafer.ai/v1";
 const WAFER_MAX_TOKENS_CAP = 65536;
 
 /**
- * Shared mapper for Wafer's `/v1/models` records.
+ * Mapper for Wafer Serverless `/v1/models` records.
  *
- * Wafer wraps each entry with a `wafer` envelope describing tier, capabilities,
- * and cents-per-million pricing. The mapper folds that metadata into the
- * canonical `ModelSpec<"openai-completions">` shape and applies zai-family thinking
- * compat when the entry advertises reasoning support (GLM-family on the Pass
- * SKU). Cents-per-million → dollars-per-million via /100.
+ * Wafer wraps each entry with a `wafer` envelope describing capabilities and
+ * pricing. The mapper folds that metadata into the canonical
+ * `ModelSpec<"openai-completions">` shape and applies upstream-specific thinking
+ * compat when the entry advertises reasoning support. Wafer pricing is exposed
+ * through internal wholesale units; the public Serverless rate equals
+ * `cents × 125 / 10000`.
  */
 interface WaferRecord {
 	context_length?: unknown;
@@ -1563,7 +1611,7 @@ function readWaferRecord(entry: OpenAICompatibleModelRecord): WaferRecord | unde
 }
 
 function mapWaferModel(
-	providerId: "wafer-pass" | "wafer-serverless",
+	providerId: "wafer-serverless",
 	baseUrl: string,
 	entry: OpenAICompatibleModelRecord,
 	defaults: ModelSpec<"openai-completions">,
@@ -1579,25 +1627,12 @@ function mapWaferModel(
 	);
 	const maxTokens = contextWindow !== null ? Math.min(contextWindow, WAFER_MAX_TOKENS_CAP) : null;
 	const pricing = wafer?.pricing ?? {};
-	// Wafer's `/v1/models` exposes pricing through `*_cents_per_million` fields,
-	// but the values are an internal wholesale unit, not literal cents — across
-	// every published Serverless model on wafer.ai the user-facing rate equals
-	// `cents × 125 / 10000` (i.e. wholesale × 1.25 / 100; GLM-5.1's `120` →
-	// $1.50/M, Kimi-K2.6's `88` → $1.10/M, etc.). The multiply-first form keeps
-	// the result a finite dyadic for every observed value.
-	// For the Pass SKU the per-token rate is bundled in the flat-rate
-	// subscription, so we follow the convention shared with
-	// `kimi-code`/`firepass`/`alibaba-coding-plan` and seed every Pass model with
-	// `cost: 0` regardless of what the upstream envelope says.
-	const isPassSku = providerId === "wafer-pass";
-	const cost = isPassSku
-		? { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 }
-		: {
-				input: (toPositiveNumber(pricing.input_cents_per_million, 0) * 125) / 10000,
-				output: (toPositiveNumber(pricing.output_cents_per_million, 0) * 125) / 10000,
-				cacheRead: (toPositiveNumber(pricing.cache_read_cents_per_million, 0) * 125) / 10000,
-				cacheWrite: 0,
-			};
+	const cost = {
+		input: (toPositiveNumber(pricing.input_cents_per_million, 0) * 125) / 10000,
+		output: (toPositiveNumber(pricing.output_cents_per_million, 0) * 125) / 10000,
+		cacheRead: (toPositiveNumber(pricing.cache_read_cents_per_million, 0) * 125) / 10000,
+		cacheWrite: 0,
+	};
 	const name = toModelName(wafer?.display_name, defaults.name);
 	const base: ModelSpec<"openai-completions"> = {
 		...defaults,
@@ -1643,13 +1678,12 @@ function mapWaferModel(
 	};
 }
 
-function createWaferOptions(
-	providerId: "wafer-pass" | "wafer-serverless",
-	config: WaferModelManagerConfig | undefined,
+export function waferServerlessModelManagerOptions(
+	config?: WaferModelManagerConfig,
 ): ModelManagerOptions<"openai-completions"> {
 	const apiKey = config?.apiKey;
 	const baseUrl = config?.baseUrl ?? WAFER_DEFAULT_BASE_URL;
-	const passOnly = providerId === "wafer-pass";
+	const providerId = "wafer-serverless" as const;
 	return {
 		providerId,
 		...(apiKey && {
@@ -1659,11 +1693,6 @@ function createWaferOptions(
 					provider: providerId,
 					baseUrl,
 					apiKey,
-					filterModel: entry => {
-						if (!passOnly) return true;
-						const wafer = readWaferRecord(entry);
-						return wafer?.tier === "pass_included";
-					},
 					mapModel: (entry, defaults) => mapWaferModel(providerId, baseUrl, entry, defaults),
 					fetch: config?.fetch,
 				}),
@@ -1671,18 +1700,6 @@ function createWaferOptions(
 	};
 }
 
-export function waferPassModelManagerOptions(
-	config?: WaferModelManagerConfig,
-): ModelManagerOptions<"openai-completions"> {
-	return createWaferOptions("wafer-pass", config);
-}
-
-export function waferServerlessModelManagerOptions(
-	config?: WaferModelManagerConfig,
-): ModelManagerOptions<"openai-completions"> {
-	return createWaferOptions("wafer-serverless", config);
-}
-
 // ---------------------------------------------------------------------------
 // 7. Mistral
 // ---------------------------------------------------------------------------
@@ -2448,7 +2465,10 @@ export function moonshotModelManagerOptions(
 	config?: MoonshotModelManagerConfig,
 ): ModelManagerOptions<"openai-completions"> {
 	const apiKey = config?.apiKey;
-	const baseUrl = config?.baseUrl ?? "https://api.moonshot.ai/v1";
+	// `MOONSHOT_BASE_URL` redirects discovery (and the streaming request that
+	// inherits this baseUrl) at the Kimi China platform `api.moonshot.cn`; an
+	// explicit `config.baseUrl` still wins. Mirrors LITELLM_BASE_URL/LM_STUDIO_BASE_URL. (#2883)
+	const baseUrl = config?.baseUrl ?? Bun.env.MOONSHOT_BASE_URL ?? "https://api.moonshot.ai/v1";
 	const references = createBundledReferenceMap<"openai-completions">("moonshot");
 	return {
 		providerId: "moonshot",