Merge main into model-registry

2026-06-22 14:00:18 +02:00
parent 019b022068 717a8f958a
commit abbd911693
220 changed files with 10488 additions and 4354 deletions
@@ -68,6 +68,8 @@ const KIMI_STATIC_HEADERS = {
 	"User-Agent": "KimiCLI/1.5",
 } as const;

+const MOONSHOT_CN_MIRRORED_MODEL_IDS = new Set(["kimi-k2.7-code", "kimi-k2.7-code-highspeed"]);
+
 const TOGETHER_BASE_URL = "https://api.together.ai/v1";
 const TOGETHER_BASE_COMPAT: OpenAICompletionsCompat = {
 	supportsStore: false,
@@ -121,6 +123,7 @@ const TOGETHER_TOGGLE_REASONING_LEVEL_MAP = {

 const AI_GATEWAY_MODELS_URL = "https://ai-gateway.vercel.sh/v1";
 const AI_GATEWAY_BASE_URL = "https://ai-gateway.vercel.sh";
+const VERTEX_BASE_URL = "https://{location}-aiplatform.googleapis.com";
 const NVIDIA_BASE_URL = "https://integrate.api.nvidia.com/v1";
 const NVIDIA_HEADERS = {
 	"NVCF-POLL-SECONDS": "3600",
@@ -154,6 +157,13 @@ const NVIDIA_NIM_UNSUPPORTED_MODELS = new Set([
 	"upstage/solar-10.7b-instruct",
 ]);
 const ZAI_TOOL_STREAM_UNSUPPORTED_MODELS = new Set(["glm-4.5", "glm-4.5-air", "glm-4.5-flash", "glm-4.5v"]);
+const ZAI_GLM52_THINKING_LEVEL_MAP = {
+	minimal: null,
+	low: "high",
+	medium: "high",
+	high: "high",
+	xhigh: "max",
+} as const;
 const EAGER_TOOL_INPUT_STREAMING_UNSUPPORTED_ANTHROPIC_MODELS = new Set([
 	"github-copilot:claude-haiku-4.5",
 	"github-copilot:claude-sonnet-4",
@@ -187,6 +197,23 @@ const OPENAI_RESPONSES_NONE_REASONING_MODELS = new Set([
 	"gpt-5.5",
 ]);

+const OPENCODE_OPENAI_COMPLETIONS_LONG_CACHE_RETENTION_UNSUPPORTED_MODELS = new Set([
+	"opencode:deepseek-v4-flash",
+	"opencode:deepseek-v4-pro",
+	"opencode:kimi-k2.5",
+	"opencode:kimi-k2.6",
+	"opencode:minimax-m2.7",
+	"opencode-go:kimi-k2.6",
+]);
+
+// Checked manually against the authenticated GitHub Copilot /models endpoint on 2026-06-15.
+// Keep this to narrow corrections over models.dev metadata instead of snapshotting Copilot's catalog.
+const GITHUB_COPILOT_THINKING_LEVEL_OVERRIDES = {
+	"claude-opus-4.7": { minimal: "low" },
+	"claude-opus-4.8": { minimal: "low" },
+	"claude-sonnet-4.6": { minimal: "low", xhigh: "max" },
+} satisfies Record<string, NonNullable<Model<Api>["thinkingLevelMap"]>>;
+
 function mergeThinkingLevelMap(model: Model<any>, map: NonNullable<Model<any>["thinkingLevelMap"]>): void {
 	model.thinkingLevelMap = { ...model.thinkingLevelMap, ...map };
 }
@@ -251,7 +278,8 @@ function isGemini3ProModel(modelId: string): boolean {
 }

 function isGemini3FlashModel(modelId: string): boolean {
-	return /gemini-3(?:\.\d+)?-flash/.test(modelId.toLowerCase());
+	const id = modelId.toLowerCase();
+	return /gemini-3(?:\.\d+)?-flash/.test(id) || id === "gemini-flash-latest" || id === "gemini-flash-lite-latest";
 }

 function isGemma4Model(modelId: string): boolean {
@@ -330,6 +358,15 @@ function applyThinkingLevelMetadata(model: Model<any>): void {
 	if (model.provider === "openai-codex" && supportsOpenAiXhigh(model.id)) {
 		mergeThinkingLevelMap(model, { minimal: "low" });
 	}
+	if (
+		(model.provider === "moonshotai" || model.provider === "moonshotai-cn") &&
+		(model.id === "kimi-k2.7-code" || model.id === "kimi-k2.7-code-highspeed")
+	) {
+		// Kimi K2.7 Code is always-thinking. Official docs say
+		// `thinking: { type: "disabled" }` is rejected, and callers can omit
+		// the thinking parameter to use the enabled default.
+		mergeThinkingLevelMap(model, { off: null });
+	}
 	if (model.provider === "openrouter" && model.id.startsWith("inception/mercury-2")) {
 		// Mercury 2 in instant mode (reasoning_effort: "none") disables tool calling.
 		// Mark "off" unsupported so the openai-completions provider omits the reasoning param
@@ -337,6 +374,12 @@ function applyThinkingLevelMetadata(model: Model<any>): void {
 		// Pi's low/medium/high pass through verbatim; OpenRouter normalizes to Mercury's vocabulary.
 		mergeThinkingLevelMap(model, { off: null });
 	}
+	if (model.provider === "openrouter" && model.id === "z-ai/glm-5.2") {
+		mergeThinkingLevelMap(model, { xhigh: "xhigh" });
+	}
+	if (model.provider === "fireworks" && model.id === "accounts/fireworks/models/glm-5p2") {
+		mergeThinkingLevelMap(model, { off: "none", minimal: null, low: "high", medium: "high", xhigh: "max" });
+	}
 	if (model.provider === "opencode-go" && model.id === "kimi-k2.6") {
 		// OpenCode Go exposes Kimi K2.6 thinking as on/off, not distinct effort tiers.
 		mergeThinkingLevelMap(model, { minimal: null, low: null, medium: null });
@@ -349,6 +392,12 @@ function applyThinkingLevelMetadata(model: Model<any>): void {
 		// Ring reasons by default. Only high/xhigh have documented explicit effort controls.
 		mergeThinkingLevelMap(model, ANT_LING_RING_THINKING_LEVEL_MAP);
 	}
+	if (model.provider === "github-copilot") {
+		const override = GITHUB_COPILOT_THINKING_LEVEL_OVERRIDES[model.id];
+		if (override) {
+			mergeThinkingLevelMap(model, override);
+		}
+	}
 }

 function getAnthropicMessagesCompat(provider: string, modelId: string): AnthropicMessagesCompat | undefined {
@@ -372,6 +421,10 @@ function normalizeNvidiaModelId(modelId: string): string {
 	return modelId.toLowerCase().replaceAll("_", ".");
 }

+function roundCost(value: number): number {
+	return Number(value.toFixed(6));
+}
+
 async function fetchNvidiaNimModelIds(): Promise<Map<string, string>> {
 	try {
 		console.log("Fetching models from NVIDIA NIM API...");
@@ -417,10 +470,10 @@ async function fetchOpenRouterModels(): Promise<Model<any>[]> {
 			}

 			// Convert pricing from $/token to $/million tokens
-			const inputCost = parseFloat(model.pricing?.prompt || "0") * 1_000_000;
-			const outputCost = parseFloat(model.pricing?.completion || "0") * 1_000_000;
-			const cacheReadCost = parseFloat(model.pricing?.input_cache_read || "0") * 1_000_000;
-			const cacheWriteCost = parseFloat(model.pricing?.input_cache_write || "0") * 1_000_000;
+			const inputCost = roundCost(parseFloat(model.pricing?.prompt || "0") * 1_000_000);
+			const outputCost = roundCost(parseFloat(model.pricing?.completion || "0") * 1_000_000);
+			const cacheReadCost = roundCost(parseFloat(model.pricing?.input_cache_read || "0") * 1_000_000);
+			const cacheWriteCost = roundCost(parseFloat(model.pricing?.input_cache_write || "0") * 1_000_000);

 			const normalizedModel: Model<any> = {
 				id: modelKey,
@@ -476,10 +529,10 @@ async function fetchAiGatewayModels(): Promise<Model<any>[]> {
 				input.push("image");
 			}

-			const inputCost = toNumber(model.pricing?.input) * 1_000_000;
-			const outputCost = toNumber(model.pricing?.output) * 1_000_000;
-			const cacheReadCost = toNumber(model.pricing?.input_cache_read) * 1_000_000;
-			const cacheWriteCost = toNumber(model.pricing?.input_cache_write) * 1_000_000;
+			const inputCost = roundCost(toNumber(model.pricing?.input) * 1_000_000);
+			const outputCost = roundCost(toNumber(model.pricing?.output) * 1_000_000);
+			const cacheReadCost = roundCost(toNumber(model.pricing?.input_cache_read) * 1_000_000);
+			const cacheWriteCost = roundCost(toNumber(model.pricing?.input_cache_write) * 1_000_000);

 			models.push({
 				id: model.id,
@@ -586,6 +639,13 @@ async function loadModelsDevData(): Promise<Model<any>[]> {
 			for (const [modelId, model] of Object.entries(data.google.models)) {
 				const m = model as ModelsDevModel;
 				if (m.tool_call !== true) continue;
+				let source = m;
+				if (modelId === "gemini-flash-latest") {
+					source = (data.google.models["gemini-3.5-flash"] as ModelsDevModel | undefined) ?? m;
+				}
+				if (modelId === "gemini-flash-lite-latest") {
+					source = (data.google.models["gemini-3.1-flash-lite"] as ModelsDevModel | undefined) ?? m;
+				}

 				models.push({
 					id: modelId,
@@ -593,16 +653,57 @@ async function loadModelsDevData(): Promise<Model<any>[]> {
 					api: "google-generative-ai",
 					provider: "google",
 					baseUrl: "https://generativelanguage.googleapis.com/v1beta",
-					reasoning: m.reasoning === true,
-					input: m.modalities?.input?.includes("image") ? ["text", "image"] : ["text"],
+					reasoning: source.reasoning === true,
+					input: source.modalities?.input?.includes("image") ? ["text", "image"] : ["text"],
 					cost: {
-						input: m.cost?.input || 0,
-						output: m.cost?.output || 0,
-						cacheRead: m.cost?.cache_read || 0,
-						cacheWrite: m.cost?.cache_write || 0,
+						input: source.cost?.input || 0,
+						output: source.cost?.output || 0,
+						cacheRead: source.cost?.cache_read || 0,
+						cacheWrite: source.cost?.cache_write || 0,
 					},
-					contextWindow: m.limit?.context || 4096,
-					maxTokens: m.limit?.output || 4096,
+					contextWindow: source.limit?.context || 4096,
+					maxTokens: source.limit?.output || 4096,
+				});
+			}
+		}
+
+		// Process Google Vertex Gemini models. The google-vertex models.dev catalog also includes
+		// Claude, OpenAI, and other MaaS models that do not use the @google/genai Gemini streaming
+		// path implemented by our google-vertex provider.
+		if (data["google-vertex"]?.models) {
+			for (const [modelId, model] of Object.entries(data["google-vertex"].models)) {
+				const m = model as ModelsDevModel;
+				if (m.tool_call !== true) continue;
+				if (!modelId.startsWith("gemini-")) continue;
+				if (modelId === "gemini-3.1-flash-lite-preview") continue;
+				let source = m;
+				if (modelId === "gemini-flash-latest") {
+					source = (data["google-vertex"].models["gemini-3.5-flash"] as ModelsDevModel | undefined) ?? m;
+				}
+				if (modelId === "gemini-flash-lite-latest") {
+					source = (data["google-vertex"].models["gemini-3.1-flash-lite"] as ModelsDevModel | undefined) ?? m;
+				}
+
+				// models.dev reports Vertex cache_read/cache_write values for Gemini 2.5 Flash that
+				// do not match the official Gemini API standard pricing table. pi only accounts
+				// cachedContentTokenCount as cacheRead.
+				const cacheRead = modelId === "gemini-2.5-flash" ? 0.03 : source.cost?.cache_read || 0;
+				models.push({
+					id: modelId,
+					name: m.name || modelId,
+					api: "google-vertex",
+					provider: "google-vertex",
+					baseUrl: VERTEX_BASE_URL,
+					reasoning: source.reasoning === true,
+					input: source.modalities?.input?.includes("image") ? ["text", "image"] : ["text"],
+					cost: {
+						input: source.cost?.input || 0,
+						output: source.cost?.output || 0,
+						cacheRead,
+						cacheWrite: 0,
+					},
+					contextWindow: source.limit?.context || 4096,
+					maxTokens: source.limit?.output || 4096,
 				});
 			}
 		}
@@ -806,6 +907,8 @@ async function loadModelsDevData(): Promise<Model<any>[]> {
 					if (m.tool_call !== true) continue;
 					const supportsImage = m.modalities?.input?.includes("image");

+					const isGlm52 = modelId === "glm-5.2";
+
 					models.push({
 						id: modelId,
 						name: m.name || modelId,
@@ -813,6 +916,7 @@ async function loadModelsDevData(): Promise<Model<any>[]> {
 						provider,
 						baseUrl,
 						reasoning: m.reasoning === true,
+						...(isGlm52 ? { thinkingLevelMap: ZAI_GLM52_THINKING_LEVEL_MAP } : {}),
 						input: supportsImage ? ["text", "image"] : ["text"],
 						cost: {
 							input: m.cost?.input || 0,
@@ -823,6 +927,7 @@ async function loadModelsDevData(): Promise<Model<any>[]> {
 						compat: {
 							supportsDeveloperRole: false,
 							thinkingFormat: "zai",
+							...(isGlm52 ? { supportsReasoningEffort: true } : {}),
 							...(!ZAI_TOOL_STREAM_UNSUPPORTED_MODELS.has(modelId) ? { zaiToolStream: true } : {}),
 						},
 						contextWindow: m.limit?.context || 4096,
@@ -849,7 +954,7 @@ async function loadModelsDevData(): Promise<Model<any>[]> {
 					cost: {
 						input: m.cost?.input || 0,
 						output: m.cost?.output || 0,
-						cacheRead: m.cost?.cache_read || 0,
+						cacheRead: m.cost?.cache_read ?? (m.cost?.input ? roundCost(m.cost.input * 0.1) : 0),
 						cacheWrite: m.cost?.cache_write || 0,
 					},
 					contextWindow: m.limit?.context || 4096,
@@ -1066,6 +1171,13 @@ async function loadModelsDevData(): Promise<Model<any>[]> {

 				if (api === "openai-completions") {
 					compat = { ...(compat ?? {}), maxTokensField: "max_tokens" };
+					if (
+						OPENCODE_OPENAI_COMPLETIONS_LONG_CACHE_RETENTION_UNSUPPORTED_MODELS.has(
+							`${variant.provider}:${modelId}`,
+						)
+					) {
+						compat = { ...compat, supportsLongCacheRetention: false };
+					}
 				}

 				models.push({
@@ -1228,12 +1340,27 @@ async function loadModelsDevData(): Promise<Model<any>[]> {
 			supportsStrictMode: false,
 			thinkingFormat: "deepseek",
 		};
+		const getMoonshotProviderModels = (key: "moonshotai" | "moonshotai-cn"): Record<string, ModelsDevModel> => {
+			const providerModels = data[key]?.models as Record<string, ModelsDevModel> | undefined;
+			return providerModels ? { ...providerModels } : {};
+		};
+		const moonshotModels = {
+			moonshotai: getMoonshotProviderModels("moonshotai"),
+			"moonshotai-cn": getMoonshotProviderModels("moonshotai-cn"),
+		};
+
+		// models.dev can lag the CN catalog while the global Moonshot catalog already
+		// has the model. Mirror selected current model IDs into moonshotai-cn until
+		// upstream CN metadata catches up.
+		for (const modelId of MOONSHOT_CN_MIRRORED_MODEL_IDS) {
+			const model = moonshotModels.moonshotai[modelId];
+			if (model && !moonshotModels["moonshotai-cn"][modelId]) {
+				moonshotModels["moonshotai-cn"][modelId] = model;
+			}
+		}

 		for (const { key, provider, baseUrl } of moonshotVariants) {
-			if (!data[key]?.models) continue;
-
-			for (const [modelId, model] of Object.entries(data[key].models)) {
-				const m = model as ModelsDevModel;
+			for (const [modelId, m] of Object.entries(moonshotModels[key])) {
 				if (m.tool_call !== true) continue;

 				models.push({
@@ -1390,7 +1517,11 @@ async function generateModels() {
 			candidate.cost.output = 1.9;
 			candidate.cost.cacheRead = 0.119;
 		}
-
+		if (candidate.provider === "fireworks" && candidate.id === "accounts/fireworks/models/glm-5p2") {
+			candidate.api = "openai-completions";
+			candidate.baseUrl = "https://api.fireworks.ai/inference/v1";
+			candidate.compat = { supportsStore: false, supportsDeveloperRole: false };
+		}
 	}


@@ -1731,9 +1862,10 @@ async function generateModels() {

 	for (const candidate of allModels) {
 		if (candidate.api === "openai-completions" && candidate.id.includes("deepseek-v4")) {
+			const preservesNativeReasoningEffort = candidate.provider === "openrouter" || candidate.provider === "opencode";
 			candidate.compat = {
 				...candidate.compat,
-				...(candidate.provider === "openrouter"
+				...(preservesNativeReasoningEffort
 					? {
 							requiresReasoningContentOnAssistantMessages:
 								deepseekCompat.requiresReasoningContentOnAssistantMessages,
@@ -1908,166 +2040,31 @@ async function generateModels() {
 		});
 	}

-	const VERTEX_BASE_URL = "https://{location}-aiplatform.googleapis.com";
-	const vertexModels: Model<"google-vertex">[] = [
-		{
-			id: "gemini-3-pro-preview",
-			name: "Gemini 3 Pro Preview (Vertex)",
-			api: "google-vertex",
-			provider: "google-vertex",
-			baseUrl: VERTEX_BASE_URL,
+	// Add "fusion" alias for openrouter/fusion. OpenRouter exposes Fusion as a
+	// router alias/plugin entry point; its model metadata does not advertise
+	// tools, but the alias resolves to a concrete model that can invoke caller
+	// tools and has the openrouter:fusion server tool auto-injected.
+	if (!allModels.some(m => m.provider === "openrouter" && m.id === "openrouter/fusion")) {
+		allModels.push({
+			id: "openrouter/fusion",
+			name: "OpenRouter: Fusion",
+			api: "openai-completions",
+			provider: "openrouter",
+			baseUrl: "https://openrouter.ai/api/v1",
 			reasoning: true,
-			input: ["text", "image"],
-			cost: { input: 2, output: 12, cacheRead: 0.2, cacheWrite: 0 },
+			input: ["text"],
+			cost: {
+				// we dont know about the costs because Fusion routes to multiple models
+				// and then charges you for the underlying used models
+				input: 0,
+				output: 0,
+				cacheRead: 0,
+				cacheWrite: 0,
+			},
 			contextWindow: 1000000,
-			maxTokens: 64000,
-		},
-		{
-			id: "gemini-3.1-pro-preview",
-			name: "Gemini 3.1 Pro Preview (Vertex)",
-			api: "google-vertex",
-			provider: "google-vertex",
-			baseUrl: VERTEX_BASE_URL,
-			reasoning: true,
-			input: ["text", "image"],
-			cost: { input: 2, output: 12, cacheRead: 0.2, cacheWrite: 0 },
-			contextWindow: 1048576,
-			maxTokens: 65536,
-		},
-		{
-			id: "gemini-3.1-pro-preview-customtools",
-			name: "Gemini 3.1 Pro Preview Custom Tools (Vertex)",
-			api: "google-vertex",
-			provider: "google-vertex",
-			baseUrl: VERTEX_BASE_URL,
-			reasoning: true,
-			input: ["text", "image"],
-			cost: { input: 2, output: 12, cacheRead: 0.2, cacheWrite: 0 },
-			contextWindow: 1048576,
-			maxTokens: 65536,
-		},
-		{
-			id: "gemini-3-flash-preview",
-			name: "Gemini 3 Flash Preview (Vertex)",
-			api: "google-vertex",
-			provider: "google-vertex",
-			baseUrl: VERTEX_BASE_URL,
-			reasoning: true,
-			input: ["text", "image"],
-			cost: { input: 0.5, output: 3, cacheRead: 0.05, cacheWrite: 0 },
-			contextWindow: 1048576,
-			maxTokens: 65536,
-		},
-		{
-			id: "gemini-2.0-flash",
-			name: "Gemini 2.0 Flash (Vertex)",
-			api: "google-vertex",
-			provider: "google-vertex",
-			baseUrl: VERTEX_BASE_URL,
-			reasoning: false,
-			input: ["text", "image"],
-			cost: { input: 0.15, output: 0.6, cacheRead: 0.0375, cacheWrite: 0 },
-			contextWindow: 1048576,
-			maxTokens: 8192,
-		},
-		{
-			id: "gemini-2.0-flash-lite",
-			name: "Gemini 2.0 Flash Lite (Vertex)",
-			api: "google-vertex",
-			provider: "google-vertex",
-			baseUrl: VERTEX_BASE_URL,
-			reasoning: true,
-			input: ["text", "image"],
-			cost: { input: 0.075, output: 0.3, cacheRead: 0.01875, cacheWrite: 0 },
-			contextWindow: 1048576,
-			maxTokens: 65536,
-		},
-		{
-			id: "gemini-2.5-pro",
-			name: "Gemini 2.5 Pro (Vertex)",
-			api: "google-vertex",
-			provider: "google-vertex",
-			baseUrl: VERTEX_BASE_URL,
-			reasoning: true,
-			input: ["text", "image"],
-			cost: { input: 1.25, output: 10, cacheRead: 0.125, cacheWrite: 0 },
-			contextWindow: 1048576,
-			maxTokens: 65536,
-		},
-		{
-			id: "gemini-2.5-flash",
-			name: "Gemini 2.5 Flash (Vertex)",
-			api: "google-vertex",
-			provider: "google-vertex",
-			baseUrl: VERTEX_BASE_URL,
-			reasoning: true,
-			input: ["text", "image"],
-			cost: { input: 0.3, output: 2.5, cacheRead: 0.03, cacheWrite: 0 },
-			contextWindow: 1048576,
-			maxTokens: 65536,
-		},
-		{
-			id: "gemini-2.5-flash-lite-preview-09-2025",
-			name: "Gemini 2.5 Flash Lite Preview 09-25 (Vertex)",
-			api: "google-vertex",
-			provider: "google-vertex",
-			baseUrl: VERTEX_BASE_URL,
-			reasoning: true,
-			input: ["text", "image"],
-			cost: { input: 0.1, output: 0.4, cacheRead: 0.01, cacheWrite: 0 },
-			contextWindow: 1048576,
-			maxTokens: 65536,
-		},
-		{
-			id: "gemini-2.5-flash-lite",
-			name: "Gemini 2.5 Flash Lite (Vertex)",
-			api: "google-vertex",
-			provider: "google-vertex",
-			baseUrl: VERTEX_BASE_URL,
-			reasoning: true,
-			input: ["text", "image"],
-			cost: { input: 0.1, output: 0.4, cacheRead: 0.01, cacheWrite: 0 },
-			contextWindow: 1048576,
-			maxTokens: 65536,
-		},
-		{
-			id: "gemini-1.5-pro",
-			name: "Gemini 1.5 Pro (Vertex)",
-			api: "google-vertex",
-			provider: "google-vertex",
-			baseUrl: VERTEX_BASE_URL,
-			reasoning: false,
-			input: ["text", "image"],
-			cost: { input: 1.25, output: 5, cacheRead: 0.3125, cacheWrite: 0 },
-			contextWindow: 1000000,
-			maxTokens: 8192,
-		},
-		{
-			id: "gemini-1.5-flash",
-			name: "Gemini 1.5 Flash (Vertex)",
-			api: "google-vertex",
-			provider: "google-vertex",
-			baseUrl: VERTEX_BASE_URL,
-			reasoning: false,
-			input: ["text", "image"],
-			cost: { input: 0.075, output: 0.3, cacheRead: 0.01875, cacheWrite: 0 },
-			contextWindow: 1000000,
-			maxTokens: 8192,
-		},
-		{
-			id: "gemini-1.5-flash-8b",
-			name: "Gemini 1.5 Flash-8B (Vertex)",
-			api: "google-vertex",
-			provider: "google-vertex",
-			baseUrl: VERTEX_BASE_URL,
-			reasoning: false,
-			input: ["text", "image"],
-			cost: { input: 0.0375, output: 0.15, cacheRead: 0.01, cacheWrite: 0 },
-			contextWindow: 1000000,
-			maxTokens: 8192,
-		},
-	];
-	allModels.push(...vertexModels);
+			maxTokens: 30000,
+		});
+	}

 	// Azure Foundry deploys these with larger context windows than OpenAI's own API,
 	// which caps gpt-5.4/gpt-5.5 at 272k. See models-sold-directly-by-azure docs.