Specify reasoning effort for OpenRouter reasoning models

Author: cte

src/api/index.ts

@@ -88,21 +88,25 @@ export function getModelParams({
 	model,
 	defaultMaxTokens,
 	defaultTemperature = 0,
+	defaultReasoningEffort,
 }: {
 	options: ApiHandlerOptions
 	model: ModelInfo
 	defaultMaxTokens?: number
 	defaultTemperature?: number
+	defaultReasoningEffort?: "low" | "medium" | "high"
 }) {
 	const {
 		modelMaxTokens: customMaxTokens,
 		modelMaxThinkingTokens: customMaxThinkingTokens,
 		modelTemperature: customTemperature,
+		reasoningEffort: customReasoningEffort,
 	} = options
 
 	let maxTokens = model.maxTokens ?? defaultMaxTokens
 	let thinking: BetaThinkingConfigParam | undefined = undefined
 	let temperature = customTemperature ?? defaultTemperature
+	const reasoningEffort = customReasoningEffort ?? defaultReasoningEffort
 
 	if (model.thinking) {
 		// Only honor `customMaxTokens` for thinking models.
@@ -118,5 +122,5 @@ export function getModelParams({
 		temperature = 1.0
 	}
 
-	return { maxTokens, thinking, temperature }
+	return { maxTokens, thinking, temperature, reasoningEffort }
 }

src/api/providers/openrouter.ts

@@ -1,8 +1,7 @@
 import { Anthropic } from "@anthropic-ai/sdk"
 import { BetaThinkingConfigParam } from "@anthropic-ai/sdk/resources/beta"
-import axios, { AxiosRequestConfig } from "axios"
+import axios from "axios"
 import OpenAI from "openai"
-import delay from "delay"
 
 import { ApiHandlerOptions, ModelInfo, openRouterDefaultModelId, openRouterDefaultModelInfo } from "../../shared/api"
 import { parseApiPrice } from "../../utils/cost"
@@ -22,6 +21,12 @@ type OpenRouterChatCompletionParams = OpenAI.Chat.ChatCompletionCreateParams & {
 	transforms?: string[]
 	include_reasoning?: boolean
 	thinking?: BetaThinkingConfigParam
+	// https://openrouter.ai/docs/use-cases/reasoning-tokens
+	reasoning?: {
+		effort?: "high" | "medium" | "low"
+		max_tokens?: number
+		exclude?: boolean
+	}
 }
 
 export class OpenRouterHandler extends BaseProvider implements SingleCompletionHandler {
@@ -42,7 +47,7 @@ export class OpenRouterHandler extends BaseProvider implements SingleCompletionH
 		systemPrompt: string,
 		messages: Anthropic.Messages.MessageParam[],
 	): AsyncGenerator<ApiStreamChunk> {
-		let { id: modelId, maxTokens, thinking, temperature, topP } = this.getModel()
+		let { id: modelId, maxTokens, thinking, temperature, topP, reasoningEffort } = this.getModel()
 
 		// Convert Anthropic messages to OpenAI format.
 		let openAiMessages: OpenAI.Chat.ChatCompletionMessageParam[] = [
@@ -70,13 +75,16 @@ export class OpenRouterHandler extends BaseProvider implements SingleCompletionH
 						},
 					],
 				}
+
 				// Add cache_control to the last two user messages
 				// (note: this works because we only ever add one user message at a time, but if we added multiple we'd need to mark the user message before the last assistant message)
 				const lastTwoUserMessages = openAiMessages.filter((msg) => msg.role === "user").slice(-2)
+
 				lastTwoUserMessages.forEach((msg) => {
 					if (typeof msg.content === "string") {
 						msg.content = [{ type: "text", text: msg.content }]
 					}
+
 					if (Array.isArray(msg.content)) {
 						// NOTE: this is fine since env details will always be added at the end. but if it weren't there, and the user added a image_url type message, it would pop a text part before it and then move it after to the end.
 						let lastTextPart = msg.content.filter((part) => part.type === "text").pop()
@@ -113,6 +121,7 @@ export class OpenRouterHandler extends BaseProvider implements SingleCompletionH
 				}),
 			// This way, the transforms field will only be included in the parameters when openRouterUseMiddleOutTransform is true.
 			...((this.options.openRouterUseMiddleOutTransform ?? true) && { transforms: ["middle-out"] }),
+			...(reasoningEffort && { reasoning: { effort: reasoningEffort } }),
 		}
 
 		const stream = await this.client.chat.completions.create(completionParams)

src/exports/roo-code.d.ts

@@ -175,10 +175,11 @@ type ProviderSettings = {
 				cachableFields?: string[] | undefined
 		  } | null)
 		| undefined
-	modelTemperature?: (number | null) | undefined
 	modelMaxTokens?: number | undefined
 	modelMaxThinkingTokens?: number | undefined
 	includeMaxTokens?: boolean | undefined
+	modelTemperature?: (number | null) | undefined
+	reasoningEffort?: ("low" | "medium" | "high") | undefined
 	rateLimitSeconds?: number | undefined
 	fakeAi?: unknown | undefined
 }

src/exports/types.ts

@@ -176,10 +176,11 @@ type ProviderSettings = {
 				cachableFields?: string[] | undefined
 		  } | null)
 		| undefined
-	modelTemperature?: (number | null) | undefined
 	modelMaxTokens?: number | undefined
 	modelMaxThinkingTokens?: number | undefined
 	includeMaxTokens?: boolean | undefined
+	modelTemperature?: (number | null) | undefined
+	reasoningEffort?: ("low" | "medium" | "high") | undefined
 	rateLimitSeconds?: number | undefined
 	fakeAi?: unknown | undefined
 }

src/schemas/index.ts

@@ -383,11 +383,12 @@ export const providerSettingsSchema = z.object({
 	requestyModelId: z.string().optional(),
 	requestyModelInfo: modelInfoSchema.nullish(),
 	// Claude 3.7 Sonnet Thinking
-	modelTemperature: z.number().nullish(),
 	modelMaxTokens: z.number().optional(),
 	modelMaxThinkingTokens: z.number().optional(),
 	// Generic
 	includeMaxTokens: z.boolean().optional(),
+	modelTemperature: z.number().nullish(),
+	reasoningEffort: z.enum(["low", "medium", "high"]).optional(),
 	rateLimitSeconds: z.number().optional(),
 	// Fake AI
 	fakeAi: z.unknown().optional(),
@@ -470,11 +471,12 @@ const providerSettingsRecord: ProviderSettingsRecord = {
 	requestyModelId: undefined,
 	requestyModelInfo: undefined,
 	// Claude 3.7 Sonnet Thinking
-	modelTemperature: undefined,
 	modelMaxTokens: undefined,
 	modelMaxThinkingTokens: undefined,
 	// Generic
 	includeMaxTokens: undefined,
+	modelTemperature: undefined,
+	reasoningEffort: undefined,
 	rateLimitSeconds: undefined,
 	// Fake AI
 	fakeAi: undefined,

webview-ui/src/components/settings/ApiOptions.tsx

@@ -46,7 +46,7 @@ import {
 	OPENROUTER_DEFAULT_PROVIDER_NAME,
 } from "@/components/ui/hooks/useOpenRouterModelProviders"
 import { Select, SelectContent, SelectItem, SelectTrigger, SelectValue, SelectSeparator, Button } from "@/components/ui"
-import { MODELS_BY_PROVIDER, PROVIDERS, VERTEX_REGIONS } from "./constants"
+import { MODELS_BY_PROVIDER, PROVIDERS, VERTEX_REGIONS, REASONING_MODELS } from "./constants"
 import { AWS_REGIONS } from "../../../../src/shared/aws_regions"
 import { VSCodeButtonLink } from "../common/VSCodeButtonLink"
 import { ModelInfoView } from "./ModelInfoView"
@@ -58,6 +58,7 @@ import { ThinkingBudget } from "./ThinkingBudget"
 import { R1FormatSetting } from "./R1FormatSetting"
 import { OpenRouterBalanceDisplay } from "./OpenRouterBalanceDisplay"
 import { RequestyBalanceDisplay } from "./RequestyBalanceDisplay"
+import { ReasoningEffort } from "./ReasoningEffort"
 
 interface ApiOptionsProps {
 	uriScheme: string | undefined
@@ -1519,6 +1520,10 @@ const ApiOptions = ({
 					</div>
 				)}
 
+			{selectedProvider === "openrouter" && REASONING_MODELS.has(selectedModelId) && (
+				<ReasoningEffort setApiConfigurationField={setApiConfigurationField} modelInfo={selectedModelInfo} />
+			)}
+
 			{selectedProvider === "glama" && (
 				<ModelPicker
 					apiConfiguration={apiConfiguration}
@@ -1646,12 +1651,14 @@ const ApiOptions = ({
 								})()}
 						</>
 					)}
+
 					<ModelInfoView
 						selectedModelId={selectedModelId}
 						modelInfo={selectedModelInfo}
 						isDescriptionExpanded={isDescriptionExpanded}
 						setIsDescriptionExpanded={setIsDescriptionExpanded}
 					/>
+
 					<ThinkingBudget
 						key={`${selectedProvider}-${selectedModelId}`}
 						apiConfiguration={apiConfiguration}

webview-ui/src/components/settings/ReasoningEffort.tsx

@@ -0,0 +1,45 @@
+import { useAppTranslation } from "@/i18n/TranslationContext"
+
+import { Select, SelectContent, SelectItem, SelectTrigger, SelectValue, SelectSeparator } from "@/components/ui"
+
+import { ApiConfiguration, ModelInfo } from "../../../../src/shared/api"
+
+export const EFFORTS = ["high", "medium", "low"] as const
+
+interface ReasoningEffortProps {
+	setApiConfigurationField: <K extends keyof ApiConfiguration>(field: K, value: ApiConfiguration[K]) => void
+	modelInfo: ModelInfo
+}
+
+export const ReasoningEffort = ({ setApiConfigurationField, modelInfo }: ReasoningEffortProps) => {
+	const { t } = useAppTranslation()
+
+	return (
+		<div className="flex flex-col gap-1">
+			<div className="flex justify-between items-center">
+				<label className="block font-medium mb-1">Model Reasoning Effort</label>
+			</div>
+			<Select
+				value={modelInfo.reasoningEffort}
+				onValueChange={(value) =>
+					setApiConfigurationField("openRouterModelInfo", {
+						...modelInfo,
+						reasoningEffort: value as "high" | "medium" | "low",
+					})
+				}>
+				<SelectTrigger className="w-full">
+					<SelectValue placeholder={t("settings:common.select")} />
+				</SelectTrigger>
+				<SelectContent>
+					<SelectItem value="openrouter">OpenRouter</SelectItem>
+					<SelectSeparator />
+					{EFFORTS.map((value) => (
+						<SelectItem key={value} value={value}>
+							{value}
+						</SelectItem>
+					))}
+				</SelectContent>
+			</Select>
+		</div>
+	)
+}

webview-ui/src/components/settings/constants.ts

@@ -46,3 +46,5 @@ export const VERTEX_REGIONS = [
 	{ value: "europe-west4", label: "europe-west4" },
 	{ value: "asia-southeast1", label: "asia-southeast1" },
 ]
+
+export const REASONING_MODELS = new Set(["x-ai/grok-3-mini-beta"])