Support Gemini 2.5 Flash thinking (#2752)

Author: cte

.changeset/shiny-poems-search.md

@@ -0,0 +1,5 @@
+---
+"roo-cline": patch
+---
+
+Support Gemini 2.5 Flash thinking mode

package-lock.json

@@ -405,7 +405,7 @@
 		"@anthropic-ai/vertex-sdk": "^0.7.0",
 		"@aws-sdk/client-bedrock-runtime": "^3.779.0",
 		"@google-cloud/vertexai": "^1.9.3",
-		"@google/generative-ai": "^0.18.0",
+		"@google/genai": "^0.9.0",
 		"@mistralai/mistralai": "^1.3.6",
 		"@modelcontextprotocol/sdk": "^1.7.0",
 		"@types/clone-deep": "^4.0.4",

package.json

@@ -1,45 +1,41 @@
-import { GeminiHandler } from "../gemini"
+// npx jest src/api/providers/__tests__/gemini.test.ts
+
 import { Anthropic } from "@anthropic-ai/sdk"
-import { GoogleGenerativeAI } from "@google/generative-ai"
-
-// Mock the Google Generative AI SDK
-jest.mock("@google/generative-ai", () => ({
-	GoogleGenerativeAI: jest.fn().mockImplementation(() => ({
-		getGenerativeModel: jest.fn().mockReturnValue({
-			generateContentStream: jest.fn(),
-			generateContent: jest.fn().mockResolvedValue({
-				response: {
-					text: () => "Test response",
-				},
-			}),
-		}),
-	})),
-}))
+
+import { GeminiHandler } from "../gemini"
+import { geminiDefaultModelId } from "../../../shared/api"
+
+const GEMINI_20_FLASH_THINKING_NAME = "gemini-2.0-flash-thinking-exp-1219"
 
 describe("GeminiHandler", () => {
 	let handler: GeminiHandler
 
 	beforeEach(() => {
+		// Create mock functions
+		const mockGenerateContentStream = jest.fn()
+		const mockGenerateContent = jest.fn()
+		const mockGetGenerativeModel = jest.fn()
+
 		handler = new GeminiHandler({
 			apiKey: "test-key",
-			apiModelId: "gemini-2.0-flash-thinking-exp-1219",
+			apiModelId: GEMINI_20_FLASH_THINKING_NAME,
 			geminiApiKey: "test-key",
 		})
+
+		// Replace the client with our mock
+		handler["client"] = {
+			models: {
+				generateContentStream: mockGenerateContentStream,
+				generateContent: mockGenerateContent,
+				getGenerativeModel: mockGetGenerativeModel,
+			},
+		} as any
 	})
 
 	describe("constructor", () => {
 		it("should initialize with provided config", () => {
 			expect(handler["options"].geminiApiKey).toBe("test-key")
-			expect(handler["options"].apiModelId).toBe("gemini-2.0-flash-thinking-exp-1219")
-		})
-
-		it.skip("should throw if API key is missing", () => {
-			expect(() => {
-				new GeminiHandler({
-					apiModelId: "gemini-2.0-flash-thinking-exp-1219",
-					geminiApiKey: "",
-				})
-			}).toThrow("API key is required for Google Gemini")
+			expect(handler["options"].apiModelId).toBe(GEMINI_20_FLASH_THINKING_NAME)
 		})
 	})
 
@@ -58,25 +54,15 @@ describe("GeminiHandler", () => {
 		const systemPrompt = "You are a helpful assistant"
 
 		it("should handle text messages correctly", async () => {
-			// Mock the stream response
-			const mockStream = {
-				stream: [{ text: () => "Hello" }, { text: () => " world!" }],
-				response: {
-					usageMetadata: {
-						promptTokenCount: 10,
-						candidatesTokenCount: 5,
-					},
+			// Setup the mock implementation to return an async generator
+			;(handler["client"].models.generateContentStream as jest.Mock).mockResolvedValue({
+				[Symbol.asyncIterator]: async function* () {
+					yield { text: "Hello" }
+					yield { text: " world!" }
+					yield { usageMetadata: { promptTokenCount: 10, candidatesTokenCount: 5 } }
 				},
-			}
-
-			// Setup the mock implementation
-			const mockGenerateContentStream = jest.fn().mockResolvedValue(mockStream)
-			const mockGetGenerativeModel = jest.fn().mockReturnValue({
-				generateContentStream: mockGenerateContentStream,
 			})
 
-			;(handler["client"] as any).getGenerativeModel = mockGetGenerativeModel
-
 			const stream = handler.createMessage(systemPrompt, mockMessages)
 			const chunks = []
 
@@ -100,99 +86,67 @@ describe("GeminiHandler", () => {
 				outputTokens: 5,
 			})
 
-			// Verify the model configuration
-			expect(mockGetGenerativeModel).toHaveBeenCalledWith(
-				{
-					model: "gemini-2.0-flash-thinking-exp-1219",
-					systemInstruction: systemPrompt,
-				},
-				{
-					baseUrl: undefined,
-				},
-			)
-
-			// Verify generation config
-			expect(mockGenerateContentStream).toHaveBeenCalledWith(
+			// Verify the call to generateContentStream
+			expect(handler["client"].models.generateContentStream).toHaveBeenCalledWith(
 				expect.objectContaining({
-					generationConfig: {
+					model: GEMINI_20_FLASH_THINKING_NAME,
+					config: expect.objectContaining({
 						temperature: 0,
-					},
+						systemInstruction: systemPrompt,
+					}),
 				}),
 			)
 		})
 
 		it("should handle API errors", async () => {
 			const mockError = new Error("Gemini API error")
-			const mockGenerateContentStream = jest.fn().mockRejectedValue(mockError)
-			const mockGetGenerativeModel = jest.fn().mockReturnValue({
-				generateContentStream: mockGenerateContentStream,
-			})
-
-			;(handler["client"] as any).getGenerativeModel = mockGetGenerativeModel
+			;(handler["client"].models.generateContentStream as jest.Mock).mockRejectedValue(mockError)
 
 			const stream = handler.createMessage(systemPrompt, mockMessages)
 
 			await expect(async () => {
 				for await (const chunk of stream) {
 					// Should throw before yielding any chunks
 				}
-			}).rejects.toThrow("Gemini API error")
+			}).rejects.toThrow()
 		})
 	})
 
 	describe("completePrompt", () => {
 		it("should complete prompt successfully", async () => {
-			const mockGenerateContent = jest.fn().mockResolvedValue({
-				response: {
-					text: () => "Test response",
-				},
-			})
-			const mockGetGenerativeModel = jest.fn().mockReturnValue({
-				generateContent: mockGenerateContent,
+			// Mock the response with text property
+			;(handler["client"].models.generateContent as jest.Mock).mockResolvedValue({
+				text: "Test response",
 			})
-			;(handler["client"] as any).getGenerativeModel = mockGetGenerativeModel
 
 			const result = await handler.completePrompt("Test prompt")
 			expect(result).toBe("Test response")
-			expect(mockGetGenerativeModel).toHaveBeenCalledWith(
-				{
-					model: "gemini-2.0-flash-thinking-exp-1219",
-				},
-				{
-					baseUrl: undefined,
-				},
-			)
-			expect(mockGenerateContent).toHaveBeenCalledWith({
+
+			// Verify the call to generateContent
+			expect(handler["client"].models.generateContent).toHaveBeenCalledWith({
+				model: GEMINI_20_FLASH_THINKING_NAME,
 				contents: [{ role: "user", parts: [{ text: "Test prompt" }] }],
-				generationConfig: {
+				config: {
+					httpOptions: undefined,
 					temperature: 0,
 				},
 			})
 		})
 
 		it("should handle API errors", async () => {
 			const mockError = new Error("Gemini API error")
-			const mockGenerateContent = jest.fn().mockRejectedValue(mockError)
-			const mockGetGenerativeModel = jest.fn().mockReturnValue({
-				generateContent: mockGenerateContent,
-			})
-			;(handler["client"] as any).getGenerativeModel = mockGetGenerativeModel
+			;(handler["client"].models.generateContent as jest.Mock).mockRejectedValue(mockError)
 
 			await expect(handler.completePrompt("Test prompt")).rejects.toThrow(
 				"Gemini completion error: Gemini API error",
 			)
 		})
 
 		it("should handle empty response", async () => {
-			const mockGenerateContent = jest.fn().mockResolvedValue({
-				response: {
-					text: () => "",
-				},
-			})
-			const mockGetGenerativeModel = jest.fn().mockReturnValue({
-				generateContent: mockGenerateContent,
+			// Mock the response with empty text
+			;(handler["client"].models.generateContent as jest.Mock).mockResolvedValue({
+				text: "",
 			})
-			;(handler["client"] as any).getGenerativeModel = mockGetGenerativeModel
 
 			const result = await handler.completePrompt("Test prompt")
 			expect(result).toBe("")
@@ -202,7 +156,7 @@ describe("GeminiHandler", () => {
 	describe("getModel", () => {
 		it("should return correct model info", () => {
 			const modelInfo = handler.getModel()
-			expect(modelInfo.id).toBe("gemini-2.0-flash-thinking-exp-1219")
+			expect(modelInfo.id).toBe(GEMINI_20_FLASH_THINKING_NAME)
 			expect(modelInfo.info).toBeDefined()
 			expect(modelInfo.info.maxTokens).toBe(8192)
 			expect(modelInfo.info.contextWindow).toBe(32_767)
@@ -214,7 +168,7 @@ describe("GeminiHandler", () => {
 				geminiApiKey: "test-key",
 			})
 			const modelInfo = invalidHandler.getModel()
-			expect(modelInfo.id).toBe("gemini-2.0-flash-001") // Default model
+			expect(modelInfo.id).toBe(geminiDefaultModelId) // Default model
 		})
 	})
 })

src/api/providers/__tests__/gemini.test.ts

@@ -23,6 +23,7 @@ export class AnthropicHandler extends BaseProvider implements SingleCompletionHa
 
 		const apiKeyFieldName =
 			this.options.anthropicBaseUrl && this.options.anthropicUseAuthToken ? "authToken" : "apiKey"
+
 		this.client = new Anthropic({
 			baseURL: this.options.anthropicBaseUrl || undefined,
 			[apiKeyFieldName]: this.options.apiKey,
@@ -217,10 +218,10 @@ export class AnthropicHandler extends BaseProvider implements SingleCompletionHa
 	}
 
 	async completePrompt(prompt: string) {
-		let { id: modelId, temperature } = this.getModel()
+		let { id: model, temperature } = this.getModel()
 
 		const message = await this.client.messages.create({
-			model: modelId,
+			model,
 			max_tokens: ANTHROPIC_DEFAULT_MAX_TOKENS,
 			thinking: undefined,
 			temperature,
@@ -241,16 +242,11 @@ export class AnthropicHandler extends BaseProvider implements SingleCompletionHa
 	override async countTokens(content: Array<Anthropic.Messages.ContentBlockParam>): Promise<number> {
 		try {
 			// Use the current model
-			const actualModelId = this.getModel().id
+			const { id: model } = this.getModel()
 
 			const response = await this.client.messages.countTokens({
-				model: actualModelId,
-				messages: [
-					{
-						role: "user",
-						content: content,
-					},
-				],
+				model,
+				messages: [{ role: "user", content: content }],
 			})
 
 			return response.input_tokens