Merge 46ea320821 into 14922f127e

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

@@ -84,7 +84,10 @@ export abstract class BaseOpenAiCompatibleProvider<ModelName extends string>
 				format: "openai",
 			}) ?? undefined
 
-		const temperature = this.options.modelTemperature ?? info.defaultTemperature ?? this.defaultTemperature
+		const temperature =
+			info.supportsTemperature === false
+				? undefined
+				: (this.options.modelTemperature ?? info.defaultTemperature ?? this.defaultTemperature)
 
 		const params: OpenAI.Chat.Completions.ChatCompletionCreateParamsStreaming = {
 			model,

src/api/providers/openai-compatible.ts

@@ -166,11 +166,12 @@ export abstract class OpenAICompatibleHandler extends BaseProvider implements Si
 		const aiSdkTools = convertToolsForAiSdk(openAiTools) as ToolSet | undefined
 
 		// Build the request options
+		const supportsTemp = this.config.modelInfo.supportsTemperature !== false
 		const requestOptions: Parameters<typeof streamText>[0] = {
 			model: languageModel,
 			system: systemPrompt,
 			messages: aiSdkMessages,
-			temperature: model.temperature ?? this.config.temperature ?? 0,
+			temperature: supportsTemp ? (model.temperature ?? this.config.temperature ?? 0) : undefined,
 			maxOutputTokens: this.getMaxOutputTokens(),
 			tools: aiSdkTools,
 			toolChoice: this.mapToolChoice(metadata?.tool_choice),
@@ -200,11 +201,12 @@ export abstract class OpenAICompatibleHandler extends BaseProvider implements Si
 	async completePrompt(prompt: string): Promise<string> {
 		const languageModel = this.getLanguageModel()
 
+		const supportsTemp = this.config.modelInfo.supportsTemperature !== false
 		const { text } = await generateText({
 			model: languageModel,
 			prompt,
 			maxOutputTokens: this.getMaxOutputTokens(),
-			temperature: this.config.temperature ?? 0,
+			temperature: supportsTemp ? (this.config.temperature ?? 0) : undefined,
 		})
 
 		return text

src/api/providers/openai.ts

@@ -154,7 +154,10 @@ 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),
+				temperature:
+					modelInfo.supportsTemperature === false
+						? undefined
+						: (this.options.modelTemperature ?? (deepseekReasoner ? DEEP_SEEK_DEFAULT_TEMPERATURE : 0)),
 				messages: convertedMessages,
 				stream: true as const,
 				...(isGrokXAI ? {} : { stream_options: { include_usage: true } }),

src/api/transform/__tests__/model-params.spec.ts

@@ -994,4 +994,82 @@ describe("getModelParams", () => {
 			expect(result.reasoningBudget).toBe(8192) // Default thinking tokens
 		})
 	})
+	describe("supportsTemperature", () => {
+		it("should set temperature to undefined for openai format when supportsTemperature is false", () => {
+			const model: ModelInfo = {
+				...baseModel,
+				supportsTemperature: false,
+			}
+
+			const result = getModelParams({
+				...openaiParams,
+				settings: { modelTemperature: 0.5 },
+				model,
+			})
+
+			expect(result.temperature).toBeUndefined()
+		})
+
+		it("should keep temperature for openai format when supportsTemperature is true", () => {
+			const model: ModelInfo = {
+				...baseModel,
+				supportsTemperature: true,
+			}
+
+			const result = getModelParams({
+				...openaiParams,
+				settings: { modelTemperature: 0.5 },
+				model,
+			})
+
+			expect(result.temperature).toBe(0.5)
+		})
+
+		it("should keep temperature for openai format when supportsTemperature is undefined", () => {
+			const result = getModelParams({
+				...openaiParams,
+				settings: { modelTemperature: 0.5 },
+				model: baseModel,
+			})
+
+			expect(result.temperature).toBe(0.5)
+		})
+
+		it("should set temperature to undefined for openrouter format when supportsTemperature is false", () => {
+			const model: ModelInfo = {
+				...baseModel,
+				supportsTemperature: false,
+			}
+
+			const result = getModelParams({
+				...openrouterParams,
+				settings: { modelTemperature: 0.5 },
+				model,
+			})
+
+			expect(result.temperature).toBeUndefined()
+		})
+
+		it("should still remove temperature for o1 model IDs in openai format", () => {
+			const result = getModelParams({
+				...openaiParams,
+				modelId: "o1-preview",
+				settings: {},
+				model: baseModel,
+			})
+
+			expect(result.temperature).toBeUndefined()
+		})
+
+		it("should still remove temperature for o3-mini model IDs in openai format", () => {
+			const result = getModelParams({
+				...openaiParams,
+				modelId: "o3-mini",
+				settings: {},
+				model: baseModel,
+			})
+
+			expect(result.temperature).toBeUndefined()
+		})
+	})
 })

src/api/transform/model-params.ts

@@ -153,9 +153,9 @@ export function getModelParams({
 			reasoning: getAnthropicReasoning({ model, reasoningBudget, reasoningEffort, settings }),
 		}
 	} else if (format === "openai") {
-		// Special case for o1 and o3-mini, which don't support temperature.
-		// TODO: Add a `supportsTemperature` field to the model info.
-		if (modelId.startsWith("o1") || modelId.startsWith("o3-mini")) {
+		// Omit temperature for models that don't support it (e.g. o1, o3-mini,
+		// claude-opus-4-7 proxied via OpenAI-compatible gateways).
+		if (model.supportsTemperature === false || modelId.startsWith("o1") || modelId.startsWith("o3-mini")) {
 			params.temperature = undefined
 		}
 
@@ -172,12 +172,8 @@ export function getModelParams({
 			reasoning: getGeminiReasoning({ model, reasoningBudget, reasoningEffort, settings }),
 		}
 	} else {
-		// Special case for o1-pro, which doesn't support temperature.
-		// Note that OpenRouter's `supported_parameters` field includes
-		// `temperature`, which is probably a bug.
-		// TODO: Add a `supportsTemperature` field to the model info and populate
-		// it appropriately in the OpenRouter fetcher.
-		if (modelId === "openai/o1-pro") {
+		// Omit temperature for models that don't support it.
+		if (model.supportsTemperature === false || modelId === "openai/o1-pro") {
 			params.temperature = undefined
 		}
 

webview-ui/src/components/settings/providers/OpenAICompatible.tsx

@@ -412,6 +412,32 @@ export const OpenAICompatible = ({
 					</div>
 				</div>
 
+				<div>
+					<div className="flex items-center gap-1">
+						<Checkbox
+							checked={apiConfiguration?.openAiCustomModelInfo?.supportsTemperature ?? true}
+							onChange={handleInputChange("openAiCustomModelInfo", (checked) => {
+								return {
+									...(apiConfiguration?.openAiCustomModelInfo || openAiModelInfoSaneDefaults),
+									supportsTemperature: checked,
+								}
+							})}>
+							<span className="font-medium">
+								{t("settings:providers.customModel.temperatureSupport.label")}
+							</span>
+						</Checkbox>
+						<StandardTooltip content={t("settings:providers.customModel.temperatureSupport.description")}>
+							<i
+								className="codicon codicon-info text-vscode-descriptionForeground"
+								style={{ fontSize: "12px" }}
+							/>
+						</StandardTooltip>
+					</div>
+					<div className="text-sm text-vscode-descriptionForeground pt-1">
+						{t("settings:providers.customModel.temperatureSupport.description")}
+					</div>
+				</div>
+
 				<div>
 					<VSCodeTextField
 						value={

webview-ui/src/i18n/locales/en/settings.json

@@ -564,6 +564,10 @@
 				"label": "Prompt Caching",
 				"description": "Is this model capable of caching prompts?"
 			},
+			"temperatureSupport": {
+				"label": "Supports Temperature",
+				"description": "Does this model accept a temperature parameter? Uncheck for models like Claude Opus 4.7 that have deprecated temperature."
+			},
 			"pricing": {
 				"input": {
 					"label": "Input Price",