fix: omit temperature from OpenAI-compatible requests when not explicitly set

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

@@ -497,6 +497,74 @@ describe("OpenAiHandler", () => {
 			const callArgs = mockCreate.mock.calls[0][0]
 			expect(callArgs.max_completion_tokens).toBe(4096)
 		})
+
+		it("should omit temperature when modelTemperature is not set (streaming)", async () => {
+			// When no custom temperature is configured, temperature should not be sent
+			// so that OpenAI-compatible APIs (e.g. Kimi) can use their own defaults
+			const handlerNoTemp = new OpenAiHandler({
+				...mockOptions,
+				// modelTemperature is not set (undefined)
+			})
+			const stream = handlerNoTemp.createMessage(systemPrompt, messages)
+			for await (const _chunk of stream) {
+			}
+			expect(mockCreate).toHaveBeenCalled()
+			const callArgs = mockCreate.mock.calls[0][0]
+			expect(callArgs).not.toHaveProperty("temperature")
+		})
+
+		it("should include temperature when modelTemperature is explicitly set (streaming)", async () => {
+			const handlerWithTemp = new OpenAiHandler({
+				...mockOptions,
+				modelTemperature: 0.6,
+			})
+			const stream = handlerWithTemp.createMessage(systemPrompt, messages)
+			for await (const _chunk of stream) {
+			}
+			expect(mockCreate).toHaveBeenCalled()
+			const callArgs = mockCreate.mock.calls[0][0]
+			expect(callArgs.temperature).toBe(0.6)
+		})
+
+		it("should include temperature 0 when modelTemperature is explicitly set to 0 (streaming)", async () => {
+			const handlerWithZeroTemp = new OpenAiHandler({
+				...mockOptions,
+				modelTemperature: 0,
+			})
+			const stream = handlerWithZeroTemp.createMessage(systemPrompt, messages)
+			for await (const _chunk of stream) {
+			}
+			expect(mockCreate).toHaveBeenCalled()
+			const callArgs = mockCreate.mock.calls[0][0]
+			expect(callArgs.temperature).toBe(0)
+		})
+
+		it("should omit temperature when modelTemperature is not set (non-streaming)", async () => {
+			const handlerNoTemp = new OpenAiHandler({
+				...mockOptions,
+				openAiStreamingEnabled: false,
+			})
+			const stream = handlerNoTemp.createMessage(systemPrompt, messages)
+			for await (const _chunk of stream) {
+			}
+			expect(mockCreate).toHaveBeenCalled()
+			const callArgs = mockCreate.mock.calls[0][0]
+			expect(callArgs).not.toHaveProperty("temperature")
+		})
+
+		it("should include temperature when modelTemperature is explicitly set (non-streaming)", async () => {
+			const handlerWithTemp = new OpenAiHandler({
+				...mockOptions,
+				openAiStreamingEnabled: false,
+				modelTemperature: 0.6,
+			})
+			const stream = handlerWithTemp.createMessage(systemPrompt, messages)
+			for await (const _chunk of stream) {
+			}
+			expect(mockCreate).toHaveBeenCalled()
+			const callArgs = mockCreate.mock.calls[0][0]
+			expect(callArgs.temperature).toBe(0.6)
+		})
 	})
 
 	describe("error handling", () => {
@@ -632,7 +700,6 @@ describe("OpenAiHandler", () => {
 					],
 					stream: true,
 					stream_options: { include_usage: true },
-					temperature: 0,
 					tools: undefined,
 					tool_choice: undefined,
 					parallel_tool_calls: true,

src/api/providers/openai.ts

@@ -154,7 +154,13 @@ export class OpenAiHandler extends BaseProvider implements SingleCompletionHandl
 
 			const requestOptions: OpenAI.Chat.Completions.ChatCompletionCreateParamsStreaming = {
 				model: modelId,
-				temperature: this.options.modelTemperature ?? (deepseekReasoner ? DEEP_SEEK_DEFAULT_TEMPERATURE : 0),
+				// Only include temperature when explicitly configured or for deepseek reasoner models.
+				// Omitting it lets OpenAI-compatible APIs (e.g. Kimi) use their own model-specific defaults.
+				...(this.options.modelTemperature != null
+					? { temperature: this.options.modelTemperature }
+					: deepseekReasoner
+						? { temperature: DEEP_SEEK_DEFAULT_TEMPERATURE }
+						: {}),
 				messages: convertedMessages,
 				stream: true as const,
 				...(isGrokXAI ? {} : { stream_options: { include_usage: true } }),
@@ -223,6 +229,13 @@ export class OpenAiHandler extends BaseProvider implements SingleCompletionHandl
 		} else {
 			const requestOptions: OpenAI.Chat.Completions.ChatCompletionCreateParamsNonStreaming = {
 				model: modelId,
+				// Only include temperature when explicitly configured or for deepseek reasoner models.
+				// Omitting it lets OpenAI-compatible APIs (e.g. Kimi) use their own model-specific defaults.
+				...(this.options.modelTemperature != null
+					? { temperature: this.options.modelTemperature }
+					: deepseekReasoner
+						? { temperature: DEEP_SEEK_DEFAULT_TEMPERATURE }
+						: {}),
 				messages: deepseekReasoner
 					? convertToR1Format([{ role: "user", content: systemPrompt }, ...messages])
 					: [systemMessage, ...convertToOpenAiMessages(messages)],