feat(xai): add dynamic model discovery with correct pricing

- Implemented dynamic model discovery for xAI provider using /v1/language-models endpoint
- Models are now fetched at runtime with real-time pricing and capabilities
- Fixed pricing conversion: XAI API returns fractional cents (basis points), divide by 10,000 not 100
- This fixes pricing display showing $20.00 instead of $0.20 per 1M tokens
- Removed static xAI models from MODELS_BY_PROVIDER to rely on dynamic discovery
- Enhanced error logging with detailed status and URL information
- Support dynamic model context window overrides from API
- Fixed parseApiPrice to handle zero values correctly (for free models)
- Provided complete ModelInfo fallback in useSelectedModel for UI type safety
- Added comprehensive test coverage for cost utilities and XAI fetcher
- Updated all tests to reflect correct pricing scale and dynamic model behavior

Author: hannesrudolph

packages/types/src/provider-settings.ts

@@ -21,7 +21,6 @@ import {
 	sambaNovaModels,
 	vertexModels,
 	vscodeLlmModels,
-	xaiModels,
 	internationalZAiModels,
 	minimaxModels,
 } from "./providers/index.js"
@@ -50,6 +49,7 @@ export const dynamicProviders = [
 	"glama",
 	"roo",
 	"chutes",
+	"xai",
 ] as const
 
 export type DynamicProvider = (typeof dynamicProviders)[number]
@@ -137,7 +137,6 @@ export const providerNames = [
 	"roo",
 	"sambanova",
 	"vertex",
-	"xai",
 	"zai",
 ] as const
 
@@ -354,6 +353,7 @@ const fakeAiSchema = baseProviderSettingsSchema.extend({
 
 const xaiSchema = apiModelIdProviderModelSchema.extend({
 	xaiApiKey: z.string().optional(),
+	xaiModelContextWindow: z.number().optional(),
 })
 
 const groqSchema = apiModelIdProviderModelSchema.extend({
@@ -709,7 +709,7 @@ export const MODELS_BY_PROVIDER: Record<
 		label: "VS Code LM API",
 		models: Object.keys(vscodeLlmModels),
 	},
-	xai: { id: "xai", label: "xAI (Grok)", models: Object.keys(xaiModels) },
+	xai: { id: "xai", label: "xAI (Grok)", models: [] },
 	zai: { id: "zai", label: "Zai", models: Object.keys(internationalZAiModels) },
 
 	// Dynamic providers; models pulled from remote APIs.

packages/types/src/providers/xai.ts

@@ -3,93 +3,63 @@ import type { ModelInfo } from "../model.js"
 // https://docs.x.ai/docs/api-reference
 export type XAIModelId = keyof typeof xaiModels
 
-export const xaiDefaultModelId: XAIModelId = "grok-code-fast-1"
+export const xaiDefaultModelId: XAIModelId = "grok-4-fast-reasoning"
+
+/**
+ * Partial ModelInfo for xAI static registry.
+ * Contains only fields not available from the xAI API:
+ * - contextWindow: Not provided by API
+ * - maxTokens: Not provided by API
+ * - description: User-friendly descriptions
+ * - supportsReasoningEffort: Special capability flag
+ *
+ * All other fields (pricing, supportsPromptCache, supportsImages) are fetched dynamically.
+ */
+type XAIStaticModelInfo = Pick<ModelInfo, "contextWindow" | "description"> & {
+	maxTokens?: number | null
+	supportsReasoningEffort?: boolean
+}
 
 export const xaiModels = {
 	"grok-code-fast-1": {
 		maxTokens: 16_384,
-		contextWindow: 262_144,
-		supportsImages: false,
-		supportsPromptCache: true,
-		inputPrice: 0.2,
-		outputPrice: 1.5,
-		cacheWritesPrice: 0.02,
-		cacheReadsPrice: 0.02,
+		contextWindow: 256_000,
 		description: "xAI's Grok Code Fast model with 256K context window",
 	},
-	"grok-4": {
-		maxTokens: 8192,
-		contextWindow: 256000,
-		supportsImages: true,
-		supportsPromptCache: true,
-		inputPrice: 3.0,
-		outputPrice: 15.0,
-		cacheWritesPrice: 0.75,
-		cacheReadsPrice: 0.75,
+	"grok-4-0709": {
+		maxTokens: 16_384,
+		contextWindow: 256_000,
 		description: "xAI's Grok-4 model with 256K context window",
 	},
+	"grok-4-fast-non-reasoning": {
+		maxTokens: 32_768,
+		contextWindow: 2_000_000,
+		description: "xAI's Grok-4 Fast Non-Reasoning model with 2M context window",
+	},
+	"grok-4-fast-reasoning": {
+		maxTokens: 32_768,
+		contextWindow: 2_000_000,
+		description: "xAI's Grok-4 Fast Reasoning model with 2M context window",
+	},
 	"grok-3": {
 		maxTokens: 8192,
-		contextWindow: 131072,
-		supportsImages: false,
-		supportsPromptCache: true,
-		inputPrice: 3.0,
-		outputPrice: 15.0,
-		cacheWritesPrice: 0.75,
-		cacheReadsPrice: 0.75,
+		contextWindow: 131_072,
 		description: "xAI's Grok-3 model with 128K context window",
 	},
-	"grok-3-fast": {
-		maxTokens: 8192,
-		contextWindow: 131072,
-		supportsImages: false,
-		supportsPromptCache: true,
-		inputPrice: 5.0,
-		outputPrice: 25.0,
-		cacheWritesPrice: 1.25,
-		cacheReadsPrice: 1.25,
-		description: "xAI's Grok-3 fast model with 128K context window",
-	},
 	"grok-3-mini": {
 		maxTokens: 8192,
-		contextWindow: 131072,
-		supportsImages: false,
-		supportsPromptCache: true,
-		inputPrice: 0.3,
-		outputPrice: 0.5,
-		cacheWritesPrice: 0.07,
-		cacheReadsPrice: 0.07,
+		contextWindow: 131_072,
 		description: "xAI's Grok-3 mini model with 128K context window",
 		supportsReasoningEffort: true,
 	},
-	"grok-3-mini-fast": {
-		maxTokens: 8192,
-		contextWindow: 131072,
-		supportsImages: false,
-		supportsPromptCache: true,
-		inputPrice: 0.6,
-		outputPrice: 4.0,
-		cacheWritesPrice: 0.15,
-		cacheReadsPrice: 0.15,
-		description: "xAI's Grok-3 mini fast model with 128K context window",
-		supportsReasoningEffort: true,
-	},
 	"grok-2-1212": {
 		maxTokens: 8192,
-		contextWindow: 131072,
-		supportsImages: false,
-		supportsPromptCache: false,
-		inputPrice: 2.0,
-		outputPrice: 10.0,
-		description: "xAI's Grok-2 model (version 1212) with 128K context window",
+		contextWindow: 32_768,
+		description: "xAI's Grok-2 model (version 1212) with 32K context window",
 	},
 	"grok-2-vision-1212": {
 		maxTokens: 8192,
-		contextWindow: 32768,
-		supportsImages: true,
-		supportsPromptCache: false,
-		inputPrice: 2.0,
-		outputPrice: 10.0,
+		contextWindow: 32_768,
 		description: "xAI's Grok-2 Vision model (version 1212) with image support and 32K context window",
 	},
-} as const satisfies Record<string, ModelInfo>
+} as const satisfies Record<string, XAIStaticModelInfo>

src/api/providers/__tests__/xai.spec.ts

@@ -57,7 +57,12 @@ describe("XAIHandler", () => {
 	it("should return default model when no model is specified", () => {
 		const model = handler.getModel()
 		expect(model.id).toBe(xaiDefaultModelId)
-		expect(model.info).toEqual(xaiModels[xaiDefaultModelId])
+		expect(model.info).toMatchObject({
+			contextWindow: xaiModels[xaiDefaultModelId].contextWindow,
+			maxTokens: xaiModels[xaiDefaultModelId].maxTokens,
+			description: xaiModels[xaiDefaultModelId].description,
+		})
+		expect(model.info.supportsPromptCache).toBe(false) // Placeholder until dynamic data loads
 	})
 
 	test("should return specified model when valid model is provided", () => {
@@ -66,7 +71,12 @@ describe("XAIHandler", () => {
 		const model = handlerWithModel.getModel()
 
 		expect(model.id).toBe(testModelId)
-		expect(model.info).toEqual(xaiModels[testModelId])
+		expect(model.info).toMatchObject({
+			contextWindow: xaiModels[testModelId].contextWindow,
+			maxTokens: xaiModels[testModelId].maxTokens,
+			description: xaiModels[testModelId].description,
+		})
+		expect(model.info.supportsPromptCache).toBe(false) // Placeholder until dynamic data loads
 	})
 
 	it("should include reasoning_effort parameter for mini models", async () => {
@@ -234,12 +244,13 @@ describe("XAIHandler", () => {
 
 		// Verify the usage data
 		expect(firstChunk.done).toBe(false)
-		expect(firstChunk.value).toEqual({
+		expect(firstChunk.value).toMatchObject({
 			type: "usage",
 			inputTokens: 10,
 			outputTokens: 20,
 			cacheReadTokens: 5,
 			cacheWriteTokens: 15,
+			totalCost: expect.any(Number),
 		})
 	})
 

src/api/providers/fetchers/__tests__/xai.spec.ts

@@ -0,0 +1,56 @@
+import { describe, it, expect, vi, beforeEach } from "vitest"
+import axios from "axios"
+
+vi.mock("axios")
+
+import { getXaiModels } from "../xai"
+import { xaiModels } from "@roo-code/types"
+
+describe("getXaiModels", () => {
+	const mockedAxios = axios as unknown as { get: ReturnType<typeof vi.fn> }
+
+	beforeEach(() => {
+		vi.clearAllMocks()
+	})
+
+	it("returns mapped models with pricing and modalities (augmenting static info when available)", async () => {
+		mockedAxios.get = vi.fn().mockResolvedValue({
+			data: {
+				models: [
+					{
+						id: "grok-3",
+						input_modalities: ["text"],
+						output_modalities: ["text"],
+						prompt_text_token_price: 2000, // 2000 fractional cents = $0.20 per 1M tokens
+						cached_prompt_text_token_price: 500, // 500 fractional cents = $0.05 per 1M tokens
+						completion_text_token_price: 10000, // 10000 fractional cents = $1.00 per 1M tokens
+						aliases: ["grok-3-latest"],
+					},
+				],
+			},
+		})
+
+		const result = await getXaiModels("key", "https://api.x.ai/v1")
+		expect(result["grok-3"]).toBeDefined()
+		expect(result["grok-3"]?.supportsImages).toBe(false)
+		expect(result["grok-3"]?.inputPrice).toBeCloseTo(0.2) // $0.20 per 1M tokens
+		expect(result["grok-3"]?.outputPrice).toBeCloseTo(1.0) // $1.00 per 1M tokens
+		expect(result["grok-3"]?.cacheReadsPrice).toBeCloseTo(0.05) // $0.05 per 1M tokens
+		// aliases are not added to avoid UI duplication
+		expect(result["grok-3-latest"]).toBeUndefined()
+	})
+
+	it("returns empty object on schema mismatches (graceful degradation)", async () => {
+		mockedAxios.get = vi.fn().mockResolvedValue({
+			data: { data: [{ bogus: true }] },
+		})
+		const result = await getXaiModels("key")
+		expect(result).toEqual({})
+	})
+
+	it("includes Authorization header when apiKey provided", async () => {
+		mockedAxios.get = vi.fn().mockResolvedValue({ data: { data: [] } })
+		await getXaiModels("secret")
+		expect((axios.get as any).mock.calls[0][1].headers.Authorization).toBe("Bearer secret")
+	})
+})

src/api/providers/fetchers/modelCache.ts

@@ -26,6 +26,7 @@ import { getDeepInfraModels } from "./deepinfra"
 import { getHuggingFaceModels } from "./huggingface"
 import { getRooModels } from "./roo"
 import { getChutesModels } from "./chutes"
+import { getXaiModels } from "./xai"
 
 const memoryCache = new NodeCache({ stdTTL: 5 * 60, checkperiod: 5 * 60 })
 
@@ -101,6 +102,9 @@ export const getModels = async (options: GetModelsOptions): Promise<ModelRecord>
 			case "huggingface":
 				models = await getHuggingFaceModels()
 				break
+			case "xai":
+				models = await getXaiModels(options.apiKey, options.baseUrl)
+				break
 			case "roo": {
 				// Roo Code Cloud provider requires baseUrl and optional apiKey
 				const rooBaseUrl =
@@ -121,7 +125,7 @@ export const getModels = async (options: GetModelsOptions): Promise<ModelRecord>
 		// Cache the fetched models (even if empty, to signify a successful fetch with no models).
 		memoryCache.set(provider, models)
 
-		await writeModels(provider, models).catch((err) =>
+		await writeModels(provider, models || {}).catch((err) =>
 			console.error(`[getModels] Error writing ${provider} models to file cache:`, err),
 		)