fix: support simple on/off reasoning toggle for grok-4-fast on OpenRouter

- Added special handling for x-ai/grok-4-fast model in OpenRouter parser
- Disabled reasoning effort for this model while keeping reasoning parameter
- Updated reasoning logic to handle models with simple on/off toggle
- Added comprehensive tests for the new functionality

Fixes #8709

Author: roomote

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

@@ -386,5 +386,76 @@ describe("OpenRouter API", () => {
 			expect(textResult.maxTokens).toBe(64000)
 			expect(imageResult.maxTokens).toBe(64000)
 		})
+
+		it("handles grok-4-fast reasoning correctly", () => {
+			const mockModel = {
+				name: "Grok 4 Fast",
+				context_length: 32768,
+				max_completion_tokens: 8192,
+			}
+
+			// Test with reasoning in supported parameters
+			const resultWithReasoning = parseOpenRouterModel({
+				id: "x-ai/grok-4-fast",
+				model: mockModel,
+				inputModality: ["text"],
+				outputModality: ["text"],
+				maxTokens: 8192,
+				supportedParameters: ["temperature", "max_tokens", "reasoning"],
+			})
+
+			expect(resultWithReasoning.supportsReasoningEffort).toBe(false)
+			expect(resultWithReasoning.supportedParameters).toContain("reasoning")
+			expect(resultWithReasoning.supportedParameters).toContain("temperature")
+			expect(resultWithReasoning.supportedParameters).toContain("max_tokens")
+
+			// Test without reasoning in supported parameters - should add it
+			const resultWithoutReasoning = parseOpenRouterModel({
+				id: "x-ai/grok-4-fast",
+				model: mockModel,
+				inputModality: ["text"],
+				outputModality: ["text"],
+				maxTokens: 8192,
+				supportedParameters: ["temperature", "max_tokens"],
+			})
+
+			expect(resultWithoutReasoning.supportsReasoningEffort).toBe(false)
+			expect(resultWithoutReasoning.supportedParameters).toContain("reasoning")
+			expect(resultWithoutReasoning.supportedParameters).toContain("temperature")
+			expect(resultWithoutReasoning.supportedParameters).toContain("max_tokens")
+
+			// Test with undefined supported parameters
+			const resultNoParams = parseOpenRouterModel({
+				id: "x-ai/grok-4-fast",
+				model: mockModel,
+				inputModality: ["text"],
+				outputModality: ["text"],
+				maxTokens: 8192,
+			})
+
+			expect(resultNoParams.supportsReasoningEffort).toBe(false)
+			expect(resultNoParams.supportedParameters).toContain("reasoning")
+		})
+
+		it("does not affect other models reasoning configuration", () => {
+			const mockModel = {
+				name: "Other Model",
+				context_length: 32768,
+				max_completion_tokens: 8192,
+			}
+
+			const result = parseOpenRouterModel({
+				id: "other/model",
+				model: mockModel,
+				inputModality: ["text"],
+				outputModality: ["text"],
+				maxTokens: 8192,
+				supportedParameters: ["temperature", "max_tokens", "reasoning"],
+			})
+
+			// Should not modify supportsReasoningEffort for other models
+			expect(result.supportsReasoningEffort).toBe(true)
+			expect(result.supportedParameters).toContain("reasoning")
+		})
 	})
 })

src/api/providers/fetchers/openrouter.ts

@@ -270,5 +270,16 @@ export const parseOpenRouterModel = ({
 		modelInfo.maxTokens = 32768
 	}
 
+	// Special handling for grok-4-fast on OpenRouter - it should use simple on/off reasoning toggle
+	// instead of reasoning effort levels (low/medium/high)
+	if (id === "x-ai/grok-4-fast") {
+		// Disable reasoning effort but keep reasoning in supportedParameters for on/off toggle
+		modelInfo.supportsReasoningEffort = false
+		// Ensure reasoning is in supportedParameters
+		if (!supportedParameters || !supportedParameters.includes("reasoning")) {
+			modelInfo.supportedParameters = [...(modelInfo.supportedParameters || []), "reasoning"]
+		}
+	}
+
 	return modelInfo
 }

src/api/transform/__tests__/reasoning.spec.ts

@@ -284,6 +284,89 @@ describe("reasoning.ts", () => {
 
 			expect(result).toBeUndefined()
 		})
+
+		it("should handle models with simple on/off reasoning toggle (like grok-4-fast)", () => {
+			// Model with reasoning support but no effort or budget capabilities
+			const modelWithSimpleToggle: ModelInfo = {
+				maxTokens: 8192,
+				contextWindow: 32768,
+				supportsPromptCache: false,
+				supportedParameters: ["reasoning"],
+				supportsReasoningEffort: false,
+				supportsReasoningBudget: false,
+			}
+
+			// Test with reasoning enabled (not minimal)
+			const optionsWithReasoningEnabled = {
+				...baseOptions,
+				model: modelWithSimpleToggle,
+				reasoningEffort: "high" as ReasoningEffortWithMinimal,
+			}
+
+			const resultEnabled = getOpenRouterReasoning(optionsWithReasoningEnabled)
+			expect(resultEnabled).toEqual({ exclude: false })
+
+			// Test with reasoning disabled (minimal)
+			const optionsWithReasoningDisabled = {
+				...baseOptions,
+				model: modelWithSimpleToggle,
+				reasoningEffort: "minimal" as ReasoningEffortWithMinimal,
+			}
+
+			const resultDisabled = getOpenRouterReasoning(optionsWithReasoningDisabled)
+			expect(resultDisabled).toEqual({ exclude: true })
+
+			// Test without reasoning effort setting (should enable by default)
+			const optionsWithoutEffort = {
+				...baseOptions,
+				model: modelWithSimpleToggle,
+			}
+
+			const resultDefault = getOpenRouterReasoning(optionsWithoutEffort)
+			expect(resultDefault).toEqual({ exclude: false })
+		})
+
+		it("should not apply simple toggle logic to models with effort or budget capabilities", () => {
+			// Model with reasoning effort capability
+			const modelWithEffort: ModelInfo = {
+				maxTokens: 8192,
+				contextWindow: 32768,
+				supportsPromptCache: false,
+				supportedParameters: ["reasoning"],
+				supportsReasoningEffort: true,
+			}
+
+			const optionsWithEffort = {
+				...baseOptions,
+				model: modelWithEffort,
+				settings: { reasoningEffort: "high" as ReasoningEffortWithMinimal },
+				reasoningEffort: "high" as ReasoningEffortWithMinimal,
+			}
+
+			const resultWithEffort = getOpenRouterReasoning(optionsWithEffort)
+			// Should use effort logic, not simple toggle
+			expect(resultWithEffort).toEqual({ effort: "high" })
+
+			// Model with reasoning budget capability
+			const modelWithBudget: ModelInfo = {
+				maxTokens: 8192,
+				contextWindow: 32768,
+				supportsPromptCache: false,
+				supportedParameters: ["reasoning"],
+				supportsReasoningBudget: true,
+			}
+
+			const optionsWithBudget = {
+				...baseOptions,
+				model: modelWithBudget,
+				settings: { enableReasoningEffort: true },
+				reasoningBudget: 1000,
+			}
+
+			const resultWithBudget = getOpenRouterReasoning(optionsWithBudget)
+			// Should use budget logic, not simple toggle
+			expect(resultWithBudget).toEqual({ max_tokens: 1000 })
+		})
 	})
 
 	describe("getAnthropicReasoning", () => {

src/api/transform/reasoning.ts

@@ -30,14 +30,34 @@ export const getOpenRouterReasoning = ({
 	reasoningBudget,
 	reasoningEffort,
 	settings,
-}: GetModelReasoningOptions): OpenRouterReasoningParams | undefined =>
-	shouldUseReasoningBudget({ model, settings })
-		? { max_tokens: reasoningBudget }
-		: shouldUseReasoningEffort({ model, settings })
-			? reasoningEffort
-				? { effort: reasoningEffort }
-				: undefined
-			: undefined
+}: GetModelReasoningOptions): OpenRouterReasoningParams | undefined => {
+	// Check if model should use reasoning budget
+	if (shouldUseReasoningBudget({ model, settings })) {
+		return { max_tokens: reasoningBudget }
+	}
+
+	// Check if model should use reasoning effort
+	if (shouldUseReasoningEffort({ model, settings })) {
+		return reasoningEffort ? { effort: reasoningEffort } : undefined
+	}
+
+	// Check if model supports reasoning but not effort or budget (e.g., grok-4-fast on OpenRouter)
+	// These models use a simple on/off toggle via the exclude parameter
+	if (
+		model.supportedParameters?.includes("reasoning") &&
+		!model.supportsReasoningEffort &&
+		!model.supportsReasoningBudget
+	) {
+		// If reasoning is not explicitly disabled, enable it
+		// We use exclude: false to enable reasoning, and exclude: true to disable
+		// Check both settings.reasoningEffort and the passed reasoningEffort parameter
+		const effectiveEffort = reasoningEffort || settings.reasoningEffort
+		const reasoningEnabled = effectiveEffort !== "minimal"
+		return reasoningEnabled ? { exclude: false } : { exclude: true }
+	}
+
+	return undefined
+}
 
 export const getAnthropicReasoning = ({
 	model,