Improve OpenRouter model fetching

package.json

@@ -485,6 +485,7 @@
 		"knip": "^5.44.4",
 		"lint-staged": "^15.2.11",
 		"mkdirp": "^3.0.1",
+		"nock": "^14.0.4",
 		"npm-run-all": "^4.1.5",
 		"prettier": "^3.4.2",
 		"rimraf": "^6.0.1",

src/api/providers/__tests__/openai.test.ts

@@ -1,7 +1,6 @@
 import { OpenAiHandler } from "../openai"
 import { ApiHandlerOptions } from "../../../shared/api"
 import { Anthropic } from "@anthropic-ai/sdk"
-import { DEEP_SEEK_DEFAULT_TEMPERATURE } from "../constants"
 
 // Mock OpenAI client
 const mockCreate = jest.fn()

src/api/providers/__tests__/openrouter.test.ts

@@ -55,7 +55,6 @@ describe("OpenRouterHandler", () => {
 				info: mockOptions.openRouterModelInfo,
 				maxTokens: 1000,
 				reasoning: undefined,
-				supportsPromptCache: false,
 				temperature: 0,
 				thinking: undefined,
 				topP: undefined,

src/api/providers/constants.ts

@@ -6,7 +6,3 @@ export const DEFAULT_HEADERS = {
 export const ANTHROPIC_DEFAULT_MAX_TOKENS = 8192
 
 export const DEEP_SEEK_DEFAULT_TEMPERATURE = 0.6
-
-export const AZURE_AI_INFERENCE_PATH = "/models/chat/completions"
-
-export const REASONING_MODELS = new Set(["x-ai/grok-3-mini-beta", "grok-3-mini-beta", "grok-3-mini-fast-beta"])

src/api/providers/fetchers/__tests__/openrouter.test.ts

@@ -0,0 +1,72 @@
+// npx jest src/api/providers/fetchers/__tests__/openrouter.test.ts
+
+import path from "path"
+
+import { back as nockBack } from "nock"
+
+import { PROMPT_CACHING_MODELS } from "../../../../shared/api"
+
+import { getOpenRouterModels } from "../openrouter"
+
+nockBack.fixtures = path.join(__dirname, "fixtures")
+nockBack.setMode("dryrun")
+
+describe("OpenRouter API", () => {
+	describe("getOpenRouterModels", () => {
+		it("fetches models and validates schema", async () => {
+			const { nockDone } = await nockBack("openrouter-models.json")
+
+			const models = await getOpenRouterModels()
+
+			expect(
+				Object.entries(models)
+					.filter(([_, model]) => model.supportsPromptCache)
+					.map(([id, _]) => id)
+					.sort(),
+			).toEqual(Array.from(PROMPT_CACHING_MODELS).sort())
+
+			expect(
+				Object.entries(models)
+					.filter(([_, model]) => model.supportsComputerUse)
+					.map(([id, _]) => id)
+					.sort(),
+			).toEqual([
+				"anthropic/claude-3.5-sonnet",
+				"anthropic/claude-3.5-sonnet:beta",
+				"anthropic/claude-3.7-sonnet",
+				"anthropic/claude-3.7-sonnet:beta",
+				"anthropic/claude-3.7-sonnet:thinking",
+			])
+
+			expect(models["anthropic/claude-3.7-sonnet"]).toEqual({
+				maxTokens: 8192,
+				contextWindow: 200000,
+				supportsImages: true,
+				supportsPromptCache: true,
+				inputPrice: 3,
+				outputPrice: 15,
+				cacheWritesPrice: 3.75,
+				cacheReadsPrice: 0.3,
+				description: expect.any(String),
+				thinking: false,
+				supportsComputerUse: true,
+			})
+
+			expect(models["anthropic/claude-3.7-sonnet:thinking"]).toEqual({
+				maxTokens: 128000,
+				contextWindow: 200000,
+				supportsImages: true,
+				supportsPromptCache: true,
+				inputPrice: 3,
+				outputPrice: 15,
+				cacheWritesPrice: 3.75,
+				cacheReadsPrice: 0.3,
+				description: expect.any(String),
+				thinking: true,
+				supportsComputerUse: true,
+			})
+
+			nockDone()
+		})
+	})
+})

src/api/providers/fetchers/openrouter.ts

@@ -0,0 +1,115 @@
+import axios from "axios"
+import { z } from "zod"
+
+import { ApiHandlerOptions, ModelInfo } from "../../../shared/api"
+import { parseApiPrice } from "../../../utils/cost"
+
+// https://openrouter.ai/api/v1/models
+export const openRouterModelSchema = z.object({
+	id: z.string(),
+	name: z.string(),
+	description: z.string().optional(),
+	context_length: z.number(),
+	max_completion_tokens: z.number().nullish(),
+	architecture: z
+		.object({
+			modality: z.string().nullish(),
+			tokenizer: z.string().nullish(),
+		})
+		.optional(),
+	pricing: z
+		.object({
+			prompt: z.string().nullish(),
+			completion: z.string().nullish(),
+			input_cache_write: z.string().nullish(),
+			input_cache_read: z.string().nullish(),
+		})
+		.optional(),
+	top_provider: z
+		.object({
+			max_completion_tokens: z.number().nullish(),
+		})
+		.optional(),
+})
+
+export type OpenRouterModel = z.infer<typeof openRouterModelSchema>
+
+const openRouterModelsResponseSchema = z.object({
+	data: z.array(openRouterModelSchema),
+})
+
+type OpenRouterModelsResponse = z.infer<typeof openRouterModelsResponseSchema>
+
+export async function getOpenRouterModels(options?: ApiHandlerOptions) {
+	const models: Record<string, ModelInfo> = {}
+	const baseURL = options?.openRouterBaseUrl || "https://openrouter.ai/api/v1"
+
+	try {
+		const response = await axios.get<OpenRouterModelsResponse>(`${baseURL}/models`)
+		const result = openRouterModelsResponseSchema.safeParse(response.data)
+		const rawModels = result.success ? result.data.data : response.data.data
+
+		if (!result.success) {
+			console.error("OpenRouter models response is invalid", result.error.format())
+		}
+
+		for (const rawModel of rawModels) {
+			const cacheWritesPrice = rawModel.pricing?.input_cache_write
+				? parseApiPrice(rawModel.pricing?.input_cache_write)
+				: undefined
+
+			const cacheReadsPrice = rawModel.pricing?.input_cache_read
+				? parseApiPrice(rawModel.pricing?.input_cache_read)
+				: undefined
+
+			// Disable prompt caching for Gemini models for now.
+			const supportsPromptCache = !!cacheWritesPrice && !!cacheReadsPrice && !rawModel.id.startsWith("google")
+
+			const modelInfo: ModelInfo = {
+				maxTokens: rawModel.top_provider?.max_completion_tokens,
+				contextWindow: rawModel.context_length,
+				supportsImages: rawModel.architecture?.modality?.includes("image"),
+				supportsPromptCache,
+				inputPrice: parseApiPrice(rawModel.pricing?.prompt),
+				outputPrice: parseApiPrice(rawModel.pricing?.completion),
+				cacheWritesPrice,
+				cacheReadsPrice,
+				description: rawModel.description,
+				thinking: rawModel.id === "anthropic/claude-3.7-sonnet:thinking",
+			}
+
+			// Then OpenRouter model definition doesn't give us any hints about computer use,
+			// so we need to set that manually.
+			// The ideal `maxTokens` values are model dependent, but we should probably DRY
+			// this up and use the values defined for the Anthropic providers.
+			switch (true) {
+				case rawModel.id.startsWith("anthropic/claude-3.7-sonnet"):
+					modelInfo.supportsComputerUse = true
+					modelInfo.maxTokens = rawModel.id === "anthropic/claude-3.7-sonnet:thinking" ? 128_000 : 8192
+					break
+				case rawModel.id.startsWith("anthropic/claude-3.5-sonnet-20240620"):
+					modelInfo.maxTokens = 8192
+					break
+				case rawModel.id.startsWith("anthropic/claude-3.5-sonnet"):
+					modelInfo.supportsComputerUse = true
+					modelInfo.maxTokens = 8192
+					break
+				case rawModel.id.startsWith("anthropic/claude-3-5-haiku"):
+				case rawModel.id.startsWith("anthropic/claude-3-opus"):
+				case rawModel.id.startsWith("anthropic/claude-3-haiku"):
+					modelInfo.maxTokens = 8192
+					break
+				default:
+					break
+			}
+
+			models[rawModel.id] = modelInfo
+		}
+	} catch (error) {
+		console.error(
+			`Error fetching OpenRouter models: ${JSON.stringify(error, Object.getOwnPropertyNames(error), 2)}`,
+		)
+	}
+
+	return models
+}

src/api/providers/openai.ts

@@ -15,7 +15,9 @@ import { convertToSimpleMessages } from "../transform/simple-format"
 import { ApiStream, ApiStreamUsageChunk } from "../transform/stream"
 import { BaseProvider } from "./base-provider"
 import { XmlMatcher } from "../../utils/xml-matcher"
-import { DEEP_SEEK_DEFAULT_TEMPERATURE, DEFAULT_HEADERS, AZURE_AI_INFERENCE_PATH } from "./constants"
+import { DEFAULT_HEADERS, DEEP_SEEK_DEFAULT_TEMPERATURE } from "./constants"
+
+export const AZURE_AI_INFERENCE_PATH = "/models/chat/completions"
 
 export interface OpenAiHandlerOptions extends ApiHandlerOptions {}