feat: add Kimi K2 thinking model to Fireworks AI provider (#9202)

Co-authored-by: Roo Code <roomote@roocode.com>

packages/types/src/providers/fireworks.ts

@@ -3,6 +3,7 @@ import type { ModelInfo } from "../model.js"
 export type FireworksModelId =
 	| "accounts/fireworks/models/kimi-k2-instruct"
 	| "accounts/fireworks/models/kimi-k2-instruct-0905"
+	| "accounts/fireworks/models/kimi-k2-thinking"
 	| "accounts/fireworks/models/minimax-m2"
 	| "accounts/fireworks/models/qwen3-235b-a22b-instruct-2507"
 	| "accounts/fireworks/models/qwen3-coder-480b-a35b-instruct"
@@ -43,6 +44,21 @@ export const fireworksModels = {
 		description:
 			"Kimi K2 is a state-of-the-art mixture-of-experts (MoE) language model with 32 billion activated parameters and 1 trillion total parameters. Trained with the Muon optimizer, Kimi K2 achieves exceptional performance across frontier knowledge, reasoning, and coding tasks while being meticulously optimized for agentic capabilities.",
 	},
+	"accounts/fireworks/models/kimi-k2-thinking": {
+		maxTokens: 16000,
+		contextWindow: 256000,
+		supportsImages: false,
+		supportsPromptCache: true,
+		supportsNativeTools: true,
+		supportsTemperature: true,
+		preserveReasoning: true,
+		defaultTemperature: 1.0,
+		inputPrice: 0.6,
+		outputPrice: 2.5,
+		cacheReadsPrice: 0.15,
+		description:
+			"The kimi-k2-thinking model is a general-purpose agentic reasoning model developed by Moonshot AI. Thanks to its strength in deep reasoning and multi-turn tool use, it can solve even the hardest problems.",
+	},
 	"accounts/fireworks/models/minimax-m2": {
 		maxTokens: 4096,
 		contextWindow: 204800,

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

@@ -115,6 +115,31 @@ describe("FireworksHandler", () => {
 		)
 	})
 
+	it("should return Kimi K2 Thinking model with correct configuration", () => {
+		const testModelId: FireworksModelId = "accounts/fireworks/models/kimi-k2-thinking"
+		const handlerWithModel = new FireworksHandler({
+			apiModelId: testModelId,
+			fireworksApiKey: "test-fireworks-api-key",
+		})
+		const model = handlerWithModel.getModel()
+		expect(model.id).toBe(testModelId)
+		expect(model.info).toEqual(
+			expect.objectContaining({
+				maxTokens: 16000,
+				contextWindow: 256000,
+				supportsImages: false,
+				supportsPromptCache: true,
+				supportsNativeTools: true,
+				supportsTemperature: true,
+				preserveReasoning: true,
+				defaultTemperature: 1.0,
+				inputPrice: 0.6,
+				outputPrice: 2.5,
+				cacheReadsPrice: 0.15,
+			}),
+		)
+	})
+
 	it("should return MiniMax M2 model with correct configuration", () => {
 		const testModelId: FireworksModelId = "accounts/fireworks/models/minimax-m2"
 		const handlerWithModel = new FireworksHandler({
@@ -424,16 +449,85 @@ describe("FireworksHandler", () => {
 		)
 	})
 
-	it("should use default temperature of 0.5", () => {
-		const testModelId: FireworksModelId = "accounts/fireworks/models/kimi-k2-instruct"
+	it("should use provider default temperature of 0.5 for models without defaultTemperature", async () => {
+		const modelId: FireworksModelId = "accounts/fireworks/models/kimi-k2-instruct"
 		const handlerWithModel = new FireworksHandler({
-			apiModelId: testModelId,
+			apiModelId: modelId,
 			fireworksApiKey: "test-fireworks-api-key",
 		})
-		const model = handlerWithModel.getModel()
-		// The temperature is set in the constructor as defaultTemperature: 0.5
-		// This test verifies the handler is configured with the correct default temperature
-		expect(handlerWithModel).toBeDefined()
+
+		mockCreate.mockImplementationOnce(() => ({
+			[Symbol.asyncIterator]: () => ({
+				async next() {
+					return { done: true }
+				},
+			}),
+		}))
+
+		const messageGenerator = handlerWithModel.createMessage("system", [])
+		await messageGenerator.next()
+
+		expect(mockCreate).toHaveBeenCalledWith(
+			expect.objectContaining({
+				temperature: 0.5,
+			}),
+			undefined,
+		)
+	})
+
+	it("should use model defaultTemperature (1.0) over provider default (0.5) for kimi-k2-thinking", async () => {
+		const modelId: FireworksModelId = "accounts/fireworks/models/kimi-k2-thinking"
+		const handlerWithModel = new FireworksHandler({
+			apiModelId: modelId,
+			fireworksApiKey: "test-fireworks-api-key",
+		})
+
+		mockCreate.mockImplementationOnce(() => ({
+			[Symbol.asyncIterator]: () => ({
+				async next() {
+					return { done: true }
+				},
+			}),
+		}))
+
+		const messageGenerator = handlerWithModel.createMessage("system", [])
+		await messageGenerator.next()
+
+		// Model's defaultTemperature (1.0) should take precedence over provider's default (0.5)
+		expect(mockCreate).toHaveBeenCalledWith(
+			expect.objectContaining({
+				temperature: 1.0,
+			}),
+			undefined,
+		)
+	})
+
+	it("should use user-specified temperature over model and provider defaults", async () => {
+		const modelId: FireworksModelId = "accounts/fireworks/models/kimi-k2-thinking"
+		const handlerWithModel = new FireworksHandler({
+			apiModelId: modelId,
+			fireworksApiKey: "test-fireworks-api-key",
+			modelTemperature: 0.7,
+		})
+
+		mockCreate.mockImplementationOnce(() => ({
+			[Symbol.asyncIterator]: () => ({
+				async next() {
+					return { done: true }
+				},
+			}),
+		}))
+
+		const messageGenerator = handlerWithModel.createMessage("system", [])
+		await messageGenerator.next()
+
+		// User-specified temperature should take precedence over everything
+		expect(mockCreate).toHaveBeenCalledWith(
+			expect.objectContaining({
+				temperature: 0.7,
+			}),
+			undefined,
+		)
 	})
 
 	it("should handle empty response in completePrompt", async () => {

src/api/providers/base-openai-compatible-provider.ts

@@ -84,7 +84,7 @@ export abstract class BaseOpenAiCompatibleProvider<ModelName extends string>
 				format: "openai",
 			}) ?? undefined
 
-		const temperature = this.options.modelTemperature ?? this.defaultTemperature
+		const temperature = this.options.modelTemperature ?? info.defaultTemperature ?? this.defaultTemperature
 
 		const params: OpenAI.Chat.Completions.ChatCompletionCreateParamsStreaming = {
 			model,