supports

Ensure temperature is only supplied when the model advertises support and expose reasoning/temperature capabilities when refreshing OpenRouter models so reasoning and temperature settings are handled correctly.

proto/cline/models.proto

@@ -97,6 +97,8 @@ message OpenRouterModelInfo {
   optional bool supports_global_endpoint = 11;
   repeated ModelTier tiers = 12;
   optional string name = 13;
+  optional bool supports_reasoning = 14;
+  optional bool supports_temperature = 15;
 }
 
 // Shared response message for model information

src/core/api/transform/openrouter-stream.ts

@@ -34,6 +34,10 @@ export async function createOpenRouterStream(
 		model.id = model.id.slice(0, -CLAUDE_SONNET_1M_SUFFIX.length)
 	}
 
+	const { supportsTemperature, supportsReasoning } = model.info
+	const budget_tokens = thinkingBudgetTokens || 0
+	const reasoningOn = budget_tokens !== 0
+
 	// prompt caching: https://openrouter.ai/docs/prompt-caching
 	// this was initially specifically for claude models (some models may 'support prompt caching' automatically without this)
 	// handles direct model.id match logic
@@ -124,7 +128,9 @@ export async function createOpenRouterStream(
 			maxTokens = 8_192
 			break
 	}
-
+	// NOTE: Extended thinking does not support non-1 temperature
+	// Some models do not support temperatures, some do not support 0,
+	// so only set them for models that are known to support them.
 	let temperature: number | undefined = 0
 	let topP: number | undefined
 	if (
@@ -153,21 +159,24 @@ export async function createOpenRouterStream(
 		case "anthropic/claude-3.7-sonnet:thinking":
 		case "anthropic/claude-3-7-sonnet":
 		case "anthropic/claude-3-7-sonnet:beta":
-			const budget_tokens = thinkingBudgetTokens || 0
-			const reasoningOn = budget_tokens !== 0
 			if (reasoningOn) {
-				temperature = undefined // extended thinking does not support non-1 temperature
+				temperature = undefined
 				reasoning = { max_tokens: budget_tokens }
 			}
 			break
 		default:
 			if (thinkingBudgetTokens && model.info?.thinkingConfig && thinkingBudgetTokens > 0) {
-				temperature = undefined // extended thinking does not support non-1 temperature
 				reasoning = { max_tokens: thinkingBudgetTokens }
+				temperature = undefined
 				break
 			}
 	}
 
+	// Disable temperature when reasoning is on or not supported
+	if (supportsReasoning || reasoningOn || !supportsTemperature) {
+		temperature = undefined
+	}
+
 	const providerPreferences = OPENROUTER_PROVIDER_PREFERENCES[model.id]
 	if (providerPreferences) {
 		openRouterProviderSorting = undefined
@@ -177,7 +186,7 @@ export async function createOpenRouterStream(
 	const stream = await client.chat.completions.create({
 		model: model.id,
 		max_tokens: maxTokens,
-		temperature: temperature,
+		temperature,
 		top_p: topP,
 		messages: openAiMessages,
 		stream: true,

src/core/controller/models/refreshOpenRouterModels.ts

@@ -91,7 +91,9 @@ export async function refreshOpenRouterModels(controller: Controller): Promise<R
 				return undefined
 			}
 			for (const rawModel of rawModels as OpenRouterRawModelInfo[]) {
-				const supportThinking = rawModel.supported_parameters?.some((p) => p === "include_reasoning" || p === "reasoning")
+				// Matches "include_reasoning" and "reasoning"
+				const supportsReasoning = rawModel.supported_parameters?.some((p) => p.includes("reasoning")) ?? false
+				const supportsTemperature = rawModel.supported_parameters?.includes("temperature") ?? false
 
 				const modelInfo: ModelInfo = {
 					name: rawModel.name,
@@ -99,6 +101,8 @@ export async function refreshOpenRouterModels(controller: Controller): Promise<R
 					contextWindow: rawModel.context_length ?? 0,
 					supportsImages: rawModel.architecture?.modality?.includes("image") ?? false,
 					supportsPromptCache: false,
+					supportsReasoning,
+					supportsTemperature,
 					inputPrice: parsePrice(rawModel.pricing?.prompt) ?? 0,
 					outputPrice: parsePrice(rawModel.pricing?.completion) ?? 0,
 					cacheWritesPrice: parsePrice(rawModel.pricing?.input_cache_write),
@@ -106,7 +110,7 @@ export async function refreshOpenRouterModels(controller: Controller): Promise<R
 					description: rawModel.description ?? "",
 					// If thinking is supported, set maxBudget with a default value as a placeholder
 					// to ensure it has a valid thinkingConfig that lets the application know thinking is supported.
-					thinkingConfig: supportThinking ? { maxBudget: ANTHROPIC_MAX_THINKING_BUDGET } : undefined,
+					thinkingConfig: supportsReasoning ? { maxBudget: ANTHROPIC_MAX_THINKING_BUDGET } : undefined,
 					supportsGlobalEndpoint: rawModel.supports_global_endpoint ?? undefined,
 					tiers: rawModel.tiers ?? undefined,
 				}

src/shared/api.ts

@@ -242,6 +242,8 @@ export interface ModelInfo {
 		outputPriceTiers?: PriceTier[] // Optional: Tiered output price when budget > 0
 	}
 	supportsGlobalEndpoint?: boolean // Whether the model supports a global endpoint with Vertex AI
+	supportsTemperature?: boolean // Whether the model supports temperature parameter
+	supportsReasoning?: boolean // Whether the model is a reasoning model
 	cacheWritesPrice?: number
 	cacheReadsPrice?: number
 	description?: string

src/shared/proto-conversions/models/typeConversion.ts

@@ -56,6 +56,8 @@ export function fromProtobufModelInfo(protoInfo: OpenRouterModelInfo): ModelInfo
 		thinkingConfig: convertThinkingConfig(protoInfo.thinkingConfig),
 		supportsGlobalEndpoint: protoInfo.supportsGlobalEndpoint,
 		tiers: protoInfo.tiers.length > 0 ? protoInfo.tiers : undefined,
+		supportsReasoning: protoInfo.supportsReasoning,
+		supportsTemperature: protoInfo.supportsTemperature,
 	}
 }
 
@@ -76,6 +78,8 @@ export function toProtobufModelInfo(modelInfo: ModelInfo): OpenRouterModelInfo {
 		thinkingConfig: toProtobufThinkingConfig(modelInfo.thinkingConfig),
 		supportsGlobalEndpoint: modelInfo.supportsGlobalEndpoint,
 		tiers: modelInfo.tiers || [],
+		supportsReasoning: modelInfo.supportsReasoning,
+		supportsTemperature: modelInfo.supportsTemperature,
 	})
 }