@oh-my-pi/pi-catalog 17.1.5 → 17.1.7

package/CHANGELOG.md

@@ -2,6 +2,23 @@
 
 ## [Unreleased]
 
+## [17.1.7] - 2026-07-27
+
+### Added
+
+- Added support for moonshotai/Kimi-K3 and kimi-k3-fast models
+- Added umans-kimi-k3 prerelease model configuration
+
+### Changed
+
+- Updated pricing and token limits for selected models
+
+## [17.1.6] - 2026-07-27
+
+### Added
+
+- Added SiliconFlow providers (`siliconflow`, `siliconflow-cn`) with dynamic-only OpenAI-compatible model discovery: no bundled catalog — the model list is fetched live from each region's `/v1/models` endpoint, with non-chat entries (embedding, reranker, image, audio, video) filtered out. Discovery hydrates pricing, context/output limits, and reasoning metadata from the provider's models.dev catalog at runtime (with bundled upstream references as a reasoning-only fallback for ids models.dev has not indexed), so reasoning models keep thinking enabled and sessions compact against real context windows. `SILICONFLOW_API_KEY` / `SILICONFLOW_CN_API_KEY` environment variables are wired into `getEnvApiKey`.
+
 ## [17.1.5] - 2026-07-27
 
 ### Fixed

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

@@ -323,6 +323,18 @@ export declare const CATALOG_PROVIDERS: readonly [{
     readonly catalogDiscovery: {
         readonly label: "Sakana AI";
     };
+}, {
+    readonly id: "siliconflow";
+    readonly defaultModel: "zai-org/GLM-5.1";
+    readonly envVars: readonly ["SILICONFLOW_API_KEY"];
+    readonly createModelManagerOptions: (config: ModelManagerConfig) => import("../index.js").ModelManagerOptions<"openai-completions", unknown>;
+    readonly dynamicModelsAuthoritative: true;
+}, {
+    readonly id: "siliconflow-cn";
+    readonly defaultModel: "deepseek-ai/DeepSeek-V4-Pro";
+    readonly envVars: readonly ["SILICONFLOW_CN_API_KEY"];
+    readonly createModelManagerOptions: (config: ModelManagerConfig) => import("../index.js").ModelManagerOptions<"openai-completions", unknown>;
+    readonly dynamicModelsAuthoritative: true;
 }, {
     readonly id: "synthetic";
     readonly defaultModel: "hf:zai-org/GLM-5.1";

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

@@ -166,6 +166,14 @@ export interface DeepSeekModelManagerConfig {
     fetch?: FetchImpl;
 }
 export declare function deepseekModelManagerOptions(config?: DeepSeekModelManagerConfig): ModelManagerOptions<"openai-completions">;
+export interface SiliconFlowModelManagerConfig {
+    apiKey?: string;
+    baseUrl?: string;
+    fetch?: FetchImpl;
+}
+export declare function isLikelySiliconFlowChatModelId(id: string): boolean;
+export declare function siliconflowModelManagerOptions(config?: SiliconFlowModelManagerConfig): ModelManagerOptions<"openai-completions">;
+export declare function siliconflowCnModelManagerOptions(config?: SiliconFlowModelManagerConfig): ModelManagerOptions<"openai-completions">;
 export interface ZhipuCodingPlanModelManagerConfig {
     apiKey?: string;
     baseUrl?: string;

package/package.json

@@ -1,7 +1,7 @@
 {
 	"type": "module",
 	"name": "@oh-my-pi/pi-catalog",
-	"version": "17.1.5",
+	"version": "17.1.7",
 	"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": "^2.12.1",
-		"@oh-my-pi/pi-utils": "17.1.5",
+		"@oh-my-pi/pi-utils": "17.1.7",
 		"arktype": "2.2.3",
 		"zod": "^4"
 	},
 	"devDependencies": {
-		"@oh-my-pi/pi-ai": "17.1.5",
+		"@oh-my-pi/pi-ai": "17.1.7",
 		"@types/bun": "^1.3.14"
 	},
 	"engines": {

package/src/models.json

@@ -13155,6 +13155,27 @@
 			"supportsComputerUse": false,
 			"supportsComputerUseConfig": false
 		},
+		"moonshotai/Kimi-K3": {
+			"id": "moonshotai/Kimi-K3",
+			"name": "Kimi K3",
+			"api": "openai-completions",
+			"provider": "baseten",
+			"baseUrl": "https://inference.baseten.co/v1",
+			"reasoning": false,
+			"input": [
+				"text",
+				"image"
+			],
+			"cost": {
+				"input": 3,
+				"output": 15,
+				"cacheRead": 0.3,
+				"cacheWrite": 0
+			},
+			"contextWindow": 1048576,
+			"maxTokens": 1048576,
+			"supportsComputerUse": false
+		},
 		"nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B": {
 			"id": "nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B",
 			"name": "Nemotron Ultra",
@@ -17111,6 +17132,40 @@
 			"supportsComputerUse": false,
 			"supportsComputerUseConfig": false
 		},
+		"kimi-k3": {
+			"id": "kimi-k3",
+			"name": "Kimi K3",
+			"api": "openai-completions",
+			"provider": "fireworks",
+			"baseUrl": "https://api.fireworks.ai/inference/v1",
+			"reasoning": true,
+			"input": [
+				"text",
+				"image"
+			],
+			"cost": {
+				"input": 3,
+				"output": 15,
+				"cacheRead": 0.3,
+				"cacheWrite": 0
+			},
+			"contextWindow": 1048576,
+			"maxTokens": 131072,
+			"thinking": {
+				"mode": "effort",
+				"efforts": [
+					"low",
+					"high",
+					"max"
+				],
+				"defaultLevel": "max",
+				"effortMap": {
+					"max": "max"
+				},
+				"requiresEffort": true
+			},
+			"supportsComputerUse": false
+		},
 		"minimax-m2.5": {
 			"id": "minimax-m2.5",
 			"name": "MiniMax-M2.5",
@@ -27541,7 +27596,8 @@
 			},
 			"contextWindow": null,
 			"maxTokens": null,
-			"supportsComputerUse": false
+			"supportsComputerUse": false,
+			"supportsComputerUseConfig": false
 		},
 		"liquid/lfm-2-24b-a2b": {
 			"id": "liquid/lfm-2-24b-a2b",
@@ -30167,7 +30223,8 @@
 			},
 			"contextWindow": null,
 			"maxTokens": null,
-			"supportsComputerUse": false
+			"supportsComputerUse": false,
+			"supportsComputerUseConfig": false
 		},
 		"openai/gpt-4o:extended": {
 			"id": "openai/gpt-4o:extended",
@@ -64047,6 +64104,38 @@
 				]
 			}
 		},
+		"kimi-k3": {
+			"id": "kimi-k3",
+			"name": "Kimi K3",
+			"api": "ollama-chat",
+			"provider": "ollama-cloud",
+			"baseUrl": "https://ollama.com",
+			"reasoning": true,
+			"thinking": {
+				"mode": "effort",
+				"efforts": [
+					"low",
+					"high",
+					"max"
+				],
+				"defaultLevel": "max",
+				"requiresEffort": true
+			},
+			"input": [
+				"text",
+				"image"
+			],
+			"cost": {
+				"input": 3,
+				"output": 15,
+				"cacheRead": 0.3,
+				"cacheWrite": 0
+			},
+			"contextWindow": 1048576,
+			"maxTokens": 8192,
+			"omitMaxOutputTokens": true,
+			"supportsComputerUse": false
+		},
 		"minimax-m2": {
 			"id": "minimax-m2",
 			"name": "minimax-m2",
@@ -68713,6 +68802,39 @@
 				]
 			}
 		},
+		"kimi-k3": {
+			"id": "kimi-k3",
+			"name": "Kimi K3",
+			"api": "openai-completions",
+			"provider": "opencode-zen",
+			"baseUrl": "https://opencode.ai/zen/v1",
+			"reasoning": true,
+			"input": [
+				"text",
+				"image"
+			],
+			"cost": {
+				"input": 3,
+				"output": 15,
+				"cacheRead": 0.3,
+				"cacheWrite": 0
+			},
+			"contextWindow": 1048576,
+			"maxTokens": 131072,
+			"thinking": {
+				"mode": "effort",
+				"efforts": [
+					"low",
+					"high",
+					"max"
+				],
+				"defaultLevel": "max",
+				"effortMap": {
+					"max": "max"
+				},
+				"requiresEffort": true
+			}
+		},
 		"laguna-s-2.1-free": {
 			"id": "laguna-s-2.1-free",
 			"name": "Laguna S 2.1 Free",
@@ -72047,8 +72169,8 @@
 				"image"
 			],
 			"cost": {
-				"input": 0.12,
-				"output": 0.35,
+				"input": 0.14,
+				"output": 0.42,
 				"cacheRead": 0.049999999999999996,
 				"cacheWrite": 0
 			},
@@ -74224,9 +74346,9 @@
 				"text"
 			],
 			"cost": {
-				"input": 0.6,
-				"output": 3.5999999999999996,
-				"cacheRead": 0.19999999999999998,
+				"input": 0.5,
+				"output": 2.2,
+				"cacheRead": 0.09999999999999999,
 				"cacheWrite": 0
 			},
 			"contextWindow": 512288,
@@ -75036,11 +75158,11 @@
 			"cost": {
 				"input": 1.25,
 				"output": 10,
-				"cacheRead": 0.125,
+				"cacheRead": 0.13,
 				"cacheWrite": 0
 			},
 			"contextWindow": 128000,
-			"maxTokens": 16384,
+			"maxTokens": 32000,
 			"supportsComputerUse": false,
 			"supportsComputerUseConfig": false
 		},
@@ -75503,10 +75625,10 @@
 				"image"
 			],
 			"cost": {
-				"input": 1,
-				"output": 6,
-				"cacheRead": 0.09999999999999999,
-				"cacheWrite": 1.25
+				"input": 0.5,
+				"output": 3,
+				"cacheRead": 0.049999999999999996,
+				"cacheWrite": 0.625
 			},
 			"contextWindow": 1050000,
 			"maxTokens": 128000,
@@ -75535,10 +75657,10 @@
 				"image"
 			],
 			"cost": {
-				"input": 1,
-				"output": 6,
-				"cacheRead": 0.09999999999999999,
-				"cacheWrite": 1.25
+				"input": 0.5,
+				"output": 3,
+				"cacheRead": 0.049999999999999996,
+				"cacheWrite": 0.625
 			},
 			"contextWindow": 1050000,
 			"maxTokens": 128000,
@@ -75631,10 +75753,10 @@
 				"image"
 			],
 			"cost": {
-				"input": 2.5,
-				"output": 15,
-				"cacheRead": 0.25,
-				"cacheWrite": 3.125
+				"input": 1.25,
+				"output": 7.5,
+				"cacheRead": 0.125,
+				"cacheWrite": 1.5625
 			},
 			"contextWindow": 1050000,
 			"maxTokens": 128000,
@@ -75663,10 +75785,10 @@
 				"image"
 			],
 			"cost": {
-				"input": 2.5,
-				"output": 15,
-				"cacheRead": 0.25,
-				"cacheWrite": 3.125
+				"input": 1.25,
+				"output": 7.5,
+				"cacheRead": 0.125,
+				"cacheWrite": 1.5625
 			},
 			"contextWindow": 1050000,
 			"maxTokens": 128000,
@@ -79676,9 +79798,9 @@
 				"text"
 			],
 			"cost": {
-				"input": 0.6692,
-				"output": 2.1032,
-				"cacheRead": 0.12428,
+				"input": 0.7714000000000001,
+				"output": 2.4244,
+				"cacheRead": 0.14326,
 				"cacheWrite": 0
 			},
 			"contextWindow": 1048576,
@@ -81152,6 +81274,40 @@
 				"escapeBuiltinToolNames": true
 			}
 		},
+		"umans-kimi-k3": {
+			"id": "umans-kimi-k3",
+			"name": "Umans Kimi K3 (prerelease)",
+			"api": "anthropic-messages",
+			"provider": "umans",
+			"baseUrl": "https://api.code.umans.ai",
+			"reasoning": true,
+			"thinking": {
+				"mode": "budget",
+				"efforts": [
+					"minimal",
+					"low",
+					"medium",
+					"high",
+					"xhigh"
+				]
+			},
+			"input": [
+				"text",
+				"image"
+			],
+			"cost": {
+				"input": 0,
+				"output": 0,
+				"cacheRead": 0,
+				"cacheWrite": 0
+			},
+			"contextWindow": 1048576,
+			"maxTokens": 131071,
+			"supportsComputerUse": false,
+			"compat": {
+				"escapeBuiltinToolNames": true
+			}
+		},
 		"umans-qwen3.6-35b-a3b": {
 			"id": "umans-qwen3.6-35b-a3b",
 			"name": "Umans Qwen3.6 35B A3B",
@@ -88444,6 +88600,37 @@
 			},
 			"supportsComputerUse": false
 		},
+		"moonshotai/kimi-k3-fast": {
+			"id": "moonshotai/kimi-k3-fast",
+			"name": "Kimi K3 Fast",
+			"api": "anthropic-messages",
+			"provider": "vercel-ai-gateway",
+			"baseUrl": "https://ai-gateway.vercel.sh",
+			"reasoning": true,
+			"input": [
+				"text",
+				"image"
+			],
+			"cost": {
+				"input": 4.5,
+				"output": 22.5,
+				"cacheRead": 0.44999999999999996,
+				"cacheWrite": 0
+			},
+			"contextWindow": 1000000,
+			"maxTokens": 131072,
+			"thinking": {
+				"mode": "budget",
+				"efforts": [
+					"low",
+					"high",
+					"max"
+				],
+				"defaultLevel": "max",
+				"requiresEffort": true
+			},
+			"supportsComputerUse": false
+		},
 		"nvidia/nemotron-3-nano-30b-a3b": {
 			"id": "nvidia/nemotron-3-nano-30b-a3b",
 			"name": "Nemotron 3 Nano 30B A3B",
@@ -91404,9 +91591,9 @@
 				"text"
 			],
 			"cost": {
-				"input": 1.75,
-				"output": 5.5,
-				"cacheRead": 0.325,
+				"input": 1.575,
+				"output": 4.95,
+				"cacheRead": 0.2875,
 				"cacheWrite": 0
 			},
 			"contextWindow": 1048576,

package/src/provider-models/descriptors.ts

@@ -41,6 +41,8 @@ import {
 	qianfanModelManagerOptions,
 	qwenPortalModelManagerOptions,
 	sakanaModelManagerOptions,
+	siliconflowCnModelManagerOptions,
+	siliconflowModelManagerOptions,
 	syntheticModelManagerOptions,
 	togetherModelManagerOptions,
 	umansModelManagerOptions,
@@ -369,6 +371,20 @@ export const CATALOG_PROVIDERS = [
 		dynamicModelsAuthoritative: true,
 		catalogDiscovery: { label: "Sakana AI" },
 	},
+	{
+		id: "siliconflow",
+		defaultModel: "zai-org/GLM-5.1",
+		envVars: ["SILICONFLOW_API_KEY"],
+		createModelManagerOptions: (config: ModelManagerConfig) => siliconflowModelManagerOptions(config),
+		dynamicModelsAuthoritative: true,
+	},
+	{
+		id: "siliconflow-cn",
+		defaultModel: "deepseek-ai/DeepSeek-V4-Pro",
+		envVars: ["SILICONFLOW_CN_API_KEY"],
+		createModelManagerOptions: (config: ModelManagerConfig) => siliconflowCnModelManagerOptions(config),
+		dynamicModelsAuthoritative: true,
+	},
 	{
 		id: "synthetic",
 		defaultModel: "hf:zai-org/GLM-5.1",

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

@@ -6,6 +6,7 @@ import {
 } from "../discovery/openai-compatible";
 import { Effort, THINKING_EFFORTS } from "../effort";
 import { FIREWORKS_FAST_SUFFIX, toFireworksPublicModelId } from "../fireworks-model-id";
+import { getBundledModelReferenceIndex } from "../identity/bundled";
 import {
 	anthropicModelSupportsThinking,
 	isGlmVisionModelId,
@@ -14,6 +15,7 @@ import {
 	isKimiModelId,
 	isReasoningGlmModelId,
 } from "../identity/family";
+import { resolveModelReference } from "../identity/reference";
 import type { ModelManagerOptions } from "../model-manager";
 import { getBundledModels } from "../models";
 import type { Api, FetchImpl, Model, ModelSpec, OpenAICompat, Provider, ThinkingConfig } from "../types";
@@ -91,10 +93,11 @@ function toInputCapabilities(value: unknown): ("text" | "image")[] {
 	return supportsImage ? ["text", "image"] : ["text"];
 }
 
-async function fetchModelsDevPayload(fetchImpl: FetchImpl = discoveryFetch()): Promise<unknown> {
+async function fetchModelsDevPayload(fetchImpl: FetchImpl = discoveryFetch(), signal?: AbortSignal): Promise<unknown> {
 	const response = await fetchImpl(MODELS_DEV_URL, {
 		method: "GET",
 		headers: { Accept: "application/json" },
+		signal,
 	});
 	if (!response.ok) {
 		throw new Error(`models.dev fetch failed: ${response.status}`);
@@ -1436,6 +1439,166 @@ export function deepseekModelManagerOptions(
 ): ModelManagerOptions<"openai-completions"> {
 	return createSimpleOpenAICompletionsOptions("deepseek", "https://api.deepseek.com", config);
 }
+
+// ---------------------------------------------------------------------------
+// 6.6 SiliconFlow
+// ---------------------------------------------------------------------------
+
+export interface SiliconFlowModelManagerConfig {
+	apiKey?: string;
+	baseUrl?: string;
+	fetch?: FetchImpl;
+}
+
+/**
+ * SiliconFlow's `/v1/models` lists every served model — including embeddings,
+ * rerankers, image, audio, and video generators that cannot serve chat
+ * completions — and carries no per-model type field, so non-chat entries are
+ * dropped by id to keep the picker usable.
+ */
+const SILICONFLOW_NON_CHAT_MODEL_TOKENS = [
+	"embedding",
+	"reranker",
+	"bge-",
+	"bce-",
+	"stable-diffusion",
+	"image",
+	"flux",
+	"kolors",
+	"sensevoice",
+	"cosyvoice",
+	"fish-speech",
+	"indextts",
+	"sovits",
+	"whisper",
+	"hunyuanvideo",
+	"wan2",
+	"ltx-video",
+	"speech",
+	"moderator",
+	"tts",
+] as const;
+
+export function isLikelySiliconFlowChatModelId(id: string): boolean {
+	const normalized = id.trim().toLowerCase();
+	if (!normalized) {
+		return false;
+	}
+	return !SILICONFLOW_NON_CHAT_MODEL_TOKENS.some(token => normalized.includes(token));
+}
+
+/**
+ * models.dev mappings consulted ONLY as a runtime metadata reference during
+ * dynamic discovery. They are deliberately absent from
+ * `MODELS_DEV_PROVIDER_DESCRIPTORS` so `generate-models.ts` never bundles
+ * SiliconFlow models — the live endpoint decides which models exist, while
+ * these entries hydrate the pricing, limits, and reasoning metadata that the
+ * endpoint's bare `{id}` rows do not carry. No filter: the join against live
+ * discovered ids already restricts hydration to chat models.
+ */
+const SILICONFLOW_MODELS_DEV_DESCRIPTORS: readonly ModelsDevProviderDescriptor[] = [
+	openAiCompletionsDescriptor("siliconflow", "siliconflow", "https://api.siliconflow.com/v1", {
+		filterModel: () => true,
+	}),
+	openAiCompletionsDescriptor("siliconflow-cn", "siliconflow-cn", "https://api.siliconflow.cn/v1", {
+		filterModel: () => true,
+	}),
+];
+
+const SILICONFLOW_MODELS_DEV_REFERENCE_TIMEOUT_MS = 5_000;
+
+async function loadSiliconFlowModelsDevReferences(
+	providerId: "siliconflow" | "siliconflow-cn",
+	fetchImpl?: FetchImpl,
+): Promise<Map<string, ModelSpec<"openai-completions">>> {
+	const descriptor = SILICONFLOW_MODELS_DEV_DESCRIPTORS.find(d => d.providerId === providerId);
+	if (!descriptor) {
+		return new Map();
+	}
+	try {
+		// Bounded: this enrichment is optional, so a stalled models.dev must not
+		// hold back the authoritative endpoint request that runs after it.
+		const payload = await withCatalogDiscoveryTimeout(SILICONFLOW_MODELS_DEV_REFERENCE_TIMEOUT_MS, signal =>
+			fetchModelsDevPayload(fetchImpl, signal),
+		);
+		return createModelsDevReferenceMap<"openai-completions">(
+			mapModelsDevToModels(payload as Record<string, unknown>, [descriptor]),
+		);
+	} catch {
+		return new Map();
+	}
+}
+
+function createSiliconFlowModelManagerOptions(
+	providerId: "siliconflow" | "siliconflow-cn",
+	defaultBaseUrl: string,
+	config?: SiliconFlowModelManagerConfig,
+): ModelManagerOptions<"openai-completions"> {
+	const apiKey = config?.apiKey;
+	const baseUrl = config?.baseUrl ?? defaultBaseUrl;
+	return {
+		providerId,
+		dynamicModelsAuthoritative: true,
+		...(apiKey && {
+			fetchDynamicModels: async () => {
+				const modelsDevReferences = await loadSiliconFlowModelsDevReferences(providerId, config?.fetch);
+				// Resolved here, not at options construction: walking the bundled
+				// reference index is only worth paying for when dynamic discovery
+				// actually runs, keeping the ModelManager cache fast path cheap.
+				const canonicalReferences = getBundledModelReferenceIndex();
+				return fetchOpenAICompatibleModels({
+					api: "openai-completions",
+					provider: providerId,
+					baseUrl,
+					apiKey,
+					filterModel: (_entry, model) => isLikelySiliconFlowChatModelId(model.id),
+					mapModel: (entry, defaults) => {
+						const modelsDevReference = modelsDevReferences.get(defaults.id);
+						if (modelsDevReference) {
+							return mapWithBundledReference(entry, defaults, modelsDevReference);
+						}
+						// ids missing from models.dev (new launches) still recover intrinsic
+						// capabilities and canonical limits from any bundled upstream/reseller
+						// entry — but never its pricing, which is provider-specific.
+						const canonical = resolveModelReference(defaults.id, canonicalReferences) as
+							| ModelSpec<"openai-completions">
+							| undefined;
+						if (!canonical) {
+							return defaults;
+						}
+						const contextWindow = canonical.contextWindow ?? defaults.contextWindow;
+						const maxTokens =
+							canonical.maxTokens != null && contextWindow != null
+								? Math.min(canonical.maxTokens, contextWindow)
+								: (canonical.maxTokens ?? defaults.maxTokens);
+						return {
+							...defaults,
+							name: toModelName(entry.name, canonical.name ?? defaults.name),
+							reasoning: canonical.reasoning,
+							input: canonical.input,
+							contextWindow,
+							maxTokens,
+						};
+					},
+					fetch: config?.fetch,
+				});
+			},
+		}),
+	};
+}
+
+export function siliconflowModelManagerOptions(
+	config?: SiliconFlowModelManagerConfig,
+): ModelManagerOptions<"openai-completions"> {
+	return createSiliconFlowModelManagerOptions("siliconflow", "https://api.siliconflow.com/v1", config);
+}
+
+export function siliconflowCnModelManagerOptions(
+	config?: SiliconFlowModelManagerConfig,
+): ModelManagerOptions<"openai-completions"> {
+	return createSiliconFlowModelManagerOptions("siliconflow-cn", "https://api.siliconflow.cn/v1", config);
+}
+
 // ---------------------------------------------------------------------------
 // 6.7 Zhipu Coding Plan
 // ---------------------------------------------------------------------------