fix: use extra_body for DeepSeek V3.1 Terminus reasoning control

apps/vscode-e2e/src/suite/index.ts

@@ -6,6 +6,7 @@ import * as vscode from "vscode"
 import type { RooCodeAPI } from "@roo-code/types"
 
 import { waitFor } from "./utils"
+import * as fs from "fs"
 
 export async function run() {
 	const extension = vscode.extensions.getExtension<RooCodeAPI>("RooVeterinaryInc.roo-cline")
@@ -16,6 +17,32 @@ export async function run() {
 
 	const api = extension.isActive ? extension.exports : await extension.activate()
 
+	// Ensure OPENROUTER_API_KEY is loaded from .env.local in CI (written by workflow)
+	// __dirname at runtime is apps/vscode-e2e/out/suite, so go up two levels
+	const envPath = path.resolve(__dirname, "..", "..", ".env.local")
+	if (!process.env.OPENROUTER_API_KEY && fs.existsSync(envPath)) {
+		try {
+			const content = fs.readFileSync(envPath, "utf8")
+			for (const rawLine of content.split("\n")) {
+				const line = rawLine.trim()
+				if (!line || line.startsWith("#")) continue
+				const m = /^\s*([A-Za-z_][A-Za-z0-9_]*)\s*=\s*(.*)\s*$/.exec(line)
+				if (!m) continue
+				const key: string = m[1] ?? ""
+				let val: string = m[2] ?? ""
+				// Strip surrounding quotes if present
+				if ((val.startsWith('"') && val.endsWith('"')) || (val.startsWith("'") && val.endsWith("'"))) {
+					val = val.slice(1, -1)
+				}
+				if (key && !(key in process.env)) {
+					;(process.env as Record<string, string>)[key] = val
+				}
+			}
+		} catch {
+			// ignore env load errors; tests may still pass without API calls
+		}
+	}
+
 	await api.setConfiguration({
 		apiProvider: "openrouter" as const,
 		openRouterApiKey: process.env.OPENROUTER_API_KEY!,

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

@@ -320,4 +320,157 @@ describe("OpenRouterHandler", () => {
 			await expect(handler.completePrompt("test prompt")).rejects.toThrow("Unexpected error")
 		})
 	})
+
+	describe("DeepSeek V3.1 Terminus", () => {
+		it("uses extra_body for reasoning control in createMessage", async () => {
+			const handler = new OpenRouterHandler({
+				...mockOptions,
+				openRouterModelId: "deepseek/deepseek-v3.1-terminus",
+			})
+
+			const mockStream = {
+				async *[Symbol.asyncIterator]() {
+					yield {
+						id: "test-id",
+						choices: [{ delta: { content: "test response" } }],
+					}
+				},
+			}
+
+			const mockCreate = vitest.fn().mockResolvedValue(mockStream)
+			;(OpenAI as any).prototype.chat = {
+				completions: { create: mockCreate },
+			} as any
+
+			// Test with reasoning disabled (default)
+			await handler.createMessage("test", []).next()
+
+			expect(mockCreate).toHaveBeenCalledWith(
+				expect.objectContaining({
+					model: "deepseek/deepseek-v3.1-terminus",
+					extra_body: {
+						chat_template_kwargs: {
+							thinking: false,
+						},
+					},
+				}),
+			)
+			expect(mockCreate).not.toHaveBeenCalledWith(
+				expect.objectContaining({
+					reasoning: expect.anything(),
+				}),
+			)
+		})
+
+		it("enables thinking when reasoning is requested", async () => {
+			// Mock getModels to return a model with reasoning capability
+			const { getModels } = await import("../fetchers/modelCache")
+			vitest.mocked(getModels).mockResolvedValueOnce({
+				"deepseek/deepseek-v3.1-terminus": {
+					maxTokens: 8192,
+					contextWindow: 128000,
+					supportsImages: false,
+					supportsPromptCache: false,
+					inputPrice: 0.5,
+					outputPrice: 1.5,
+					description: "DeepSeek V3.1 Terminus",
+					reasoningEffort: "high",
+				},
+			})
+
+			const handler = new OpenRouterHandler({
+				...mockOptions,
+				openRouterModelId: "deepseek/deepseek-v3.1-terminus",
+				reasoningEffort: "high",
+			})
+
+			const mockStream = {
+				async *[Symbol.asyncIterator]() {
+					yield {
+						id: "test-id",
+						choices: [{ delta: { content: "test response" } }],
+					}
+				},
+			}
+
+			const mockCreate = vitest.fn().mockResolvedValue(mockStream)
+			;(OpenAI as any).prototype.chat = {
+				completions: { create: mockCreate },
+			} as any
+
+			await handler.createMessage("test", []).next()
+
+			expect(mockCreate).toHaveBeenCalledWith(
+				expect.objectContaining({
+					model: "deepseek/deepseek-v3.1-terminus",
+					extra_body: {
+						chat_template_kwargs: {
+							thinking: true,
+						},
+					},
+				}),
+			)
+		})
+
+		it("uses extra_body for reasoning control in completePrompt", async () => {
+			const handler = new OpenRouterHandler({
+				...mockOptions,
+				openRouterModelId: "deepseek/deepseek-v3.1-terminus",
+			})
+
+			const mockResponse = { choices: [{ message: { content: "test completion" } }] }
+
+			const mockCreate = vitest.fn().mockResolvedValue(mockResponse)
+			;(OpenAI as any).prototype.chat = {
+				completions: { create: mockCreate },
+			} as any
+
+			await handler.completePrompt("test prompt")
+
+			expect(mockCreate).toHaveBeenCalledWith(
+				expect.objectContaining({
+					model: "deepseek/deepseek-v3.1-terminus",
+					extra_body: {
+						chat_template_kwargs: {
+							thinking: false,
+						},
+					},
+				}),
+			)
+			expect(mockCreate).not.toHaveBeenCalledWith(
+				expect.objectContaining({
+					reasoning: expect.anything(),
+				}),
+			)
+		})
+
+		it("does not use extra_body for other models", async () => {
+			const handler = new OpenRouterHandler({
+				...mockOptions,
+				openRouterModelId: "anthropic/claude-sonnet-4",
+			})
+
+			const mockStream = {
+				async *[Symbol.asyncIterator]() {
+					yield {
+						id: "test-id",
+						choices: [{ delta: { content: "test response" } }],
+					}
+				},
+			}
+
+			const mockCreate = vitest.fn().mockResolvedValue(mockStream)
+			;(OpenAI as any).prototype.chat = {
+				completions: { create: mockCreate },
+			} as any
+
+			await handler.createMessage("test", []).next()
+
+			expect(mockCreate).not.toHaveBeenCalledWith(
+				expect.objectContaining({
+					extra_body: expect.anything(),
+				}),
+			)
+		})
+	})
 })

src/api/providers/openrouter.ts

@@ -60,6 +60,12 @@ type OpenRouterChatCompletionParams = OpenAI.Chat.ChatCompletionCreateParams & {
 	include_reasoning?: boolean
 	// https://openrouter.ai/docs/use-cases/reasoning-tokens
 	reasoning?: OpenRouterReasoningParams
+	// For DeepSeek models that require extra_body
+	extra_body?: {
+		chat_template_kwargs?: {
+			thinking?: boolean
+		}
+	}
 }
 
 // See `OpenAI.Chat.Completions.ChatCompletionChunk["usage"]`
@@ -141,6 +147,23 @@ export class OpenRouterHandler extends BaseProvider implements SingleCompletionH
 
 		const transforms = (this.options.openRouterUseMiddleOutTransform ?? true) ? ["middle-out"] : undefined
 
+		// For DeepSeek V3.1 Terminus, use extra_body to control reasoning
+		const isDeepSeekV3Terminus = modelId === "deepseek/deepseek-v3.1-terminus"
+		let extraBody: OpenRouterChatCompletionParams["extra_body"] = undefined
+
+		if (isDeepSeekV3Terminus) {
+			// Default to reasoning OFF for DeepSeek V3.1 Terminus
+			// Enable only if reasoning is explicitly requested
+			const enableThinking = Boolean(
+				reasoning && !reasoning.exclude && (reasoning.max_tokens || reasoning.effort),
+			)
+			extraBody = {
+				chat_template_kwargs: {
+					thinking: enableThinking,
+				},
+			}
+		}
+
 		// https://openrouter.ai/docs/transforms
 		const completionParams: OpenRouterChatCompletionParams = {
 			model: modelId,
@@ -160,7 +183,8 @@ export class OpenRouterHandler extends BaseProvider implements SingleCompletionH
 					},
 				}),
 			...(transforms && { transforms }),
-			...(reasoning && { reasoning }),
+			// For DeepSeek V3.1 Terminus, use extra_body instead of reasoning param
+			...(isDeepSeekV3Terminus ? { extra_body: extraBody } : reasoning && { reasoning }),
 		}
 
 		let stream
@@ -248,6 +272,23 @@ export class OpenRouterHandler extends BaseProvider implements SingleCompletionH
 	async completePrompt(prompt: string) {
 		let { id: modelId, maxTokens, temperature, reasoning } = await this.fetchModel()
 
+		// For DeepSeek V3.1 Terminus, use extra_body to control reasoning
+		const isDeepSeekV3Terminus = modelId === "deepseek/deepseek-v3.1-terminus"
+		let extraBody: OpenRouterChatCompletionParams["extra_body"] = undefined
+
+		if (isDeepSeekV3Terminus) {
+			// Default to reasoning OFF for DeepSeek V3.1 Terminus
+			// Enable only if reasoning is explicitly requested
+			const enableThinking = Boolean(
+				reasoning && !reasoning.exclude && (reasoning.max_tokens || reasoning.effort),
+			)
+			extraBody = {
+				chat_template_kwargs: {
+					thinking: enableThinking,
+				},
+			}
+		}
+
 		const completionParams: OpenRouterChatCompletionParams = {
 			model: modelId,
 			max_tokens: maxTokens,
@@ -263,7 +304,8 @@ export class OpenRouterHandler extends BaseProvider implements SingleCompletionH
 						allow_fallbacks: false,
 					},
 				}),
-			...(reasoning && { reasoning }),
+			// For DeepSeek V3.1 Terminus, use extra_body instead of reasoning param
+			...(isDeepSeekV3Terminus ? { extra_body: extraBody } : reasoning && { reasoning }),
 		}
 
 		let response