Gemini 2.5 flash determining its own thinking

.husky/pre-push

@@ -7,7 +7,13 @@ fi
 
 # Detect if running on Windows and use pnpm.cmd, otherwise use pnpm.
 if [ "$OS" = "Windows_NT" ]; then
-  pnpm_cmd="pnpm.cmd"
+  if command -v pnpm.cmd >/dev/null 2>&1; then
+    pnpm_cmd="pnpm.cmd"
+  elif command -v pnpm >/dev/null 2>&1; then
+    pnpm_cmd="pnpm"
+  else
+    pnpm_cmd="npx pnpm"
+  fi
 else
   if command -v pnpm >/dev/null 2>&1; then
     pnpm_cmd="pnpm"

apps/vscode-e2e/package.json

@@ -2,7 +2,7 @@
 	"name": "@roo-code/vscode-e2e",
 	"private": true,
 	"scripts": {
-		"lint": "eslint src --ext=ts --max-warnings=0",
+		"lint": "eslint src --max-warnings=0",
 		"check-types": "tsc -p tsconfig.esm.json --noEmit",
 		"format": "prettier --write src",
 		"test:ci": "pnpm -w bundle && pnpm --filter @roo-code/vscode-webview build && pnpm test:run",

packages/build/package.json

@@ -6,7 +6,7 @@
 	"main": "./dist/index.js",
 	"types": "./src/index.ts",
 	"scripts": {
-		"lint": "eslint src --ext=ts --max-warnings=0",
+		"lint": "eslint src --max-warnings=0",
 		"check-types": "tsc --noEmit",
 		"test": "vitest run",
 		"build": "tsc",

packages/cloud/package.json

@@ -5,7 +5,7 @@
 	"type": "module",
 	"exports": "./src/index.ts",
 	"scripts": {
-		"lint": "eslint src --ext=ts --max-warnings=0",
+		"lint": "eslint src --max-warnings=0",
 		"check-types": "tsc --noEmit",
 		"test": "vitest run",
 		"clean": "rimraf dist .turbo"

packages/telemetry/package.json

@@ -5,7 +5,7 @@
 	"type": "module",
 	"exports": "./src/index.ts",
 	"scripts": {
-		"lint": "eslint src --ext=ts --max-warnings=0",
+		"lint": "eslint src --max-warnings=0",
 		"check-types": "tsc --noEmit",
 		"test": "vitest run",
 		"clean": "rimraf dist .turbo"

packages/types/package.json

@@ -14,7 +14,7 @@
 		}
 	},
 	"scripts": {
-		"lint": "eslint src --ext=ts --max-warnings=0",
+		"lint": "eslint src --max-warnings=0",
 		"check-types": "tsc --noEmit",
 		"test": "vitest run",
 		"build": "tsup",

packages/types/src/providers/gemini.ts

@@ -15,7 +15,7 @@ export const geminiModels = {
 		outputPrice: 3.5,
 		maxThinkingTokens: 24_576,
 		supportsReasoningBudget: true,
-		requiredReasoningBudget: true,
+		requiredReasoningBudget: false,
 	},
 	"gemini-2.5-flash-preview-04-17": {
 		maxTokens: 65_535,
@@ -36,7 +36,7 @@ export const geminiModels = {
 		cacheWritesPrice: 1.0,
 		maxThinkingTokens: 24_576,
 		supportsReasoningBudget: true,
-		requiredReasoningBudget: true,
+		requiredReasoningBudget: false,
 	},
 	"gemini-2.5-flash-preview-05-20": {
 		maxTokens: 65_535,

packages/types/src/providers/vertex.ts

@@ -15,7 +15,7 @@ export const vertexModels = {
 		outputPrice: 3.5,
 		maxThinkingTokens: 24_576,
 		supportsReasoningBudget: true,
-		requiredReasoningBudget: true,
+		requiredReasoningBudget: false,
 	},
 	"gemini-2.5-flash-preview-05-20": {
 		maxTokens: 65_535,
@@ -34,7 +34,7 @@ export const vertexModels = {
 		outputPrice: 3.5,
 		maxThinkingTokens: 24_576,
 		supportsReasoningBudget: true,
-		requiredReasoningBudget: true,
+		requiredReasoningBudget: false,
 	},
 	"gemini-2.5-flash-preview-04-17": {
 		maxTokens: 65_535,

src/api/providers/gemini.ts

@@ -81,7 +81,28 @@ export class GeminiHandler extends BaseProvider implements SingleCompletionHandl
 		let lastUsageMetadata: GenerateContentResponseUsageMetadata | undefined
 
 		for await (const chunk of result) {
-			if (chunk.text) {
+			// Process candidates and their parts to separate thoughts from content
+			if (chunk.candidates && chunk.candidates.length > 0) {
+				const candidate = chunk.candidates[0]
+				if (candidate.content && candidate.content.parts) {
+					for (const part of candidate.content.parts) {
+						if (part.thought) {
+							// This is a thinking/reasoning part
+							if (part.text) {
+								yield { type: "reasoning", text: part.text }
+							}
+						} else {
+							// This is regular content
+							if (part.text) {
+								yield { type: "text", text: part.text }
+							}
+						}
+					}
+				}
+			}
+
+			// Fallback to the original text property if no candidates structure
+			else if (chunk.text) {
 				yield { type: "text", text: chunk.text }
 			}
 
@@ -117,12 +138,14 @@ export class GeminiHandler extends BaseProvider implements SingleCompletionHandl
 			if (geminiModels[id as GeminiModelId]) {
 				info = geminiModels[id as GeminiModelId]
 
+				const thinkingConfig = this.options.modelMaxThinkingTokens
+					? { includeThoughts: true, thinkingBudget: this.options.modelMaxThinkingTokens }
+					: { includeThoughts: true }
+
 				return {
 					id,
 					info,
-					thinkingConfig: this.options.modelMaxThinkingTokens
-						? { thinkingBudget: this.options.modelMaxThinkingTokens }
-						: undefined,
+					thinkingConfig,
 					maxOutputTokens: this.options.modelMaxTokens ?? info.maxTokens ?? undefined,
 				}
 			}

src/api/providers/vertex.ts

@@ -20,12 +20,14 @@ export class VertexHandler extends GeminiHandler implements SingleCompletionHand
 			if (vertexModels[id as VertexModelId]) {
 				info = vertexModels[id as VertexModelId]
 
+				const thinkingConfig = this.options.modelMaxThinkingTokens
+					? { includeThoughts: true, thinkingBudget: this.options.modelMaxThinkingTokens }
+					: { includeThoughts: true }
+
 				return {
 					id,
 					info,
-					thinkingConfig: this.options.modelMaxThinkingTokens
-						? { thinkingBudget: this.options.modelMaxThinkingTokens }
-						: undefined,
+					thinkingConfig,
 					maxOutputTokens: this.options.modelMaxTokens ?? info.maxTokens ?? undefined,
 				}
 			}

src/package.json

@@ -339,7 +339,7 @@
 		}
 	},
 	"scripts": {
-		"lint": "eslint . --ext=ts --max-warnings=0",
+		"lint": "eslint . --max-warnings=0",
 		"check-types": "tsc --noEmit",
 		"pretest": "turbo run bundle --cwd ..",
 		"test": "jest -w=40% && vitest run",

webview-ui/eslint.config.mjs

@@ -5,6 +5,16 @@ export default [
 	...reactConfig,
 	{
 		rules: {
+			// Disable base ESLint rule to avoid conflicts with TypeScript ESLint version
+			"no-unused-expressions": "off",
+			"@typescript-eslint/no-unused-expressions": [
+				"error",
+				{
+					allowShortCircuit: true,
+					allowTernary: true,
+					allowTaggedTemplates: true,
+				},
+			],
 			"@typescript-eslint/no-unused-vars": [
 				"error",
 				{

webview-ui/package.json

@@ -3,7 +3,7 @@
 	"private": true,
 	"type": "module",
 	"scripts": {
-		"lint": "eslint src --ext=ts,tsx --max-warnings=0",
+		"lint": "eslint src --max-warnings=0",
 		"check-types": "tsc",
 		"pretest": "turbo run bundle --cwd ..",
 		"test": "jest -w=40%",

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

@@ -21,7 +21,9 @@ export const ThinkingBudget = ({ apiConfiguration, setApiConfigurationField, mod
 	const isReasoningBudgetRequired = !!modelInfo && modelInfo.requiredReasoningBudget
 	const isReasoningEffortSupported = !!modelInfo && modelInfo.supportsReasoningEffort
 
-	const enableReasoningEffort = apiConfiguration.enableReasoningEffort
+	// For Gemini Flash thinking models, default to enabling reasoning if not explicitly set
+	const isGeminiFlashThinking = modelInfo && modelInfo.supportsReasoningBudget && !modelInfo.requiredReasoningBudget
+	const enableReasoningEffort = apiConfiguration.enableReasoningEffort ?? (isGeminiFlashThinking ? true : false)
 	const customMaxOutputTokens = apiConfiguration.modelMaxTokens || DEFAULT_HYBRID_REASONING_MODEL_MAX_TOKENS
 	const customMaxThinkingTokens =
 		apiConfiguration.modelMaxThinkingTokens || DEFAULT_HYBRID_REASONING_MODEL_THINKING_TOKENS
@@ -32,6 +34,13 @@ export const ThinkingBudget = ({ apiConfiguration, setApiConfigurationField, mod
 		? Math.min(modelInfo.maxThinkingTokens, Math.floor(0.8 * customMaxOutputTokens))
 		: Math.floor(0.8 * customMaxOutputTokens)
 
+	// Initialize enableReasoningEffort to true for Gemini Flash thinking models if not set
+	useEffect(() => {
+		if (isGeminiFlashThinking && apiConfiguration.enableReasoningEffort === undefined) {
+			setApiConfigurationField("enableReasoningEffort", true)
+		}
+	}, [isGeminiFlashThinking, apiConfiguration.enableReasoningEffort, setApiConfigurationField])
+
 	// If the custom max thinking tokens are going to exceed it's limit due
 	// to the custom max output tokens being reduced then we need to shrink it
 	// appropriately.
@@ -73,19 +82,54 @@ export const ThinkingBudget = ({ apiConfiguration, setApiConfigurationField, mod
 							<div className="w-12 text-sm text-center">{customMaxOutputTokens}</div>
 						</div>
 					</div>
-					<div className="flex flex-col gap-1">
-						<div className="font-medium">{t("settings:thinkingBudget.maxThinkingTokens")}</div>
-						<div className="flex items-center gap-1" data-testid="reasoning-budget">
-							<Slider
-								min={1024}
-								max={modelMaxThinkingTokens}
-								step={1024}
-								value={[customMaxThinkingTokens]}
-								onValueChange={([value]) => setApiConfigurationField("modelMaxThinkingTokens", value)}
-							/>
-							<div className="w-12 text-sm text-center">{customMaxThinkingTokens}</div>
+					{isGeminiFlashThinking ? (
+						<div className="flex flex-col gap-1">
+							<Checkbox
+								checked={!!apiConfiguration.modelMaxThinkingTokens}
+								onChange={(checked: boolean) => {
+									if (checked) {
+										setApiConfigurationField(
+											"modelMaxThinkingTokens",
+											DEFAULT_HYBRID_REASONING_MODEL_THINKING_TOKENS,
+										)
+									} else {
+										setApiConfigurationField("modelMaxThinkingTokens", undefined)
+									}
+								}}>
+								{t("settings:thinkingBudget.manualThinkingBudget")}
+							</Checkbox>
+							{!!apiConfiguration.modelMaxThinkingTokens && (
+								<div className="flex items-center gap-1" data-testid="reasoning-budget">
+									<Slider
+										min={1024}
+										max={modelMaxThinkingTokens}
+										step={1024}
+										value={[customMaxThinkingTokens]}
+										onValueChange={([value]) =>
+											setApiConfigurationField("modelMaxThinkingTokens", value)
+										}
+									/>
+									<div className="w-12 text-sm text-center">{customMaxThinkingTokens}</div>
+								</div>
+							)}
 						</div>
-					</div>
+					) : (
+						<div className="flex flex-col gap-1">
+							<div className="font-medium">{t("settings:thinkingBudget.maxThinkingTokens")}</div>
+							<div className="flex items-center gap-1" data-testid="reasoning-budget">
+								<Slider
+									min={1024}
+									max={modelMaxThinkingTokens}
+									step={1024}
+									value={[customMaxThinkingTokens]}
+									onValueChange={([value]) =>
+										setApiConfigurationField("modelMaxThinkingTokens", value)
+									}
+								/>
+								<div className="w-12 text-sm text-center">{customMaxThinkingTokens}</div>
+							</div>
+						</div>
+					)}
 				</>
 			)}
 		</>

webview-ui/src/i18n/locales/ca/settings.json

@@ -547,7 +547,8 @@
 	},
 	"thinkingBudget": {
 		"maxTokens": "Tokens màxims",
-		"maxThinkingTokens": "Tokens de pensament màxims"
+		"maxThinkingTokens": "Tokens de pensament màxims",
+		"manualThinkingBudget": "Establir pressupost manual de pensament (si no, el model decideix automàticament)"
 	},
 	"validation": {
 		"apiKey": "Heu de proporcionar una clau API vàlida.",

webview-ui/src/i18n/locales/de/settings.json

@@ -547,7 +547,8 @@
 	},
 	"thinkingBudget": {
 		"maxTokens": "Maximale Tokens",
-		"maxThinkingTokens": "Maximale Thinking-Tokens"
+		"maxThinkingTokens": "Maximale Thinking-Tokens",
+		"manualThinkingBudget": "Manuelles Thinking-Budget festlegen (sonst entscheidet das Modell automatisch)"
 	},
 	"validation": {
 		"apiKey": "Du musst einen gültigen API-Schlüssel angeben.",

webview-ui/src/i18n/locales/en/settings.json

@@ -547,7 +547,8 @@
 	},
 	"thinkingBudget": {
 		"maxTokens": "Max Tokens",
-		"maxThinkingTokens": "Max Thinking Tokens"
+		"maxThinkingTokens": "Max Thinking Tokens",
+		"manualThinkingBudget": "Set manual thinking budget (otherwise model decides automatically)"
 	},
 	"validation": {
 		"apiKey": "You must provide a valid API key.",

webview-ui/src/i18n/locales/es/settings.json

@@ -547,7 +547,8 @@
 	},
 	"thinkingBudget": {
 		"maxTokens": "Tokens máximos",
-		"maxThinkingTokens": "Tokens máximos de pensamiento"
+		"maxThinkingTokens": "Tokens máximos de pensamiento",
+		"manualThinkingBudget": "Establecer presupuesto manual de pensamientos (de lo contrario el modelo decide automáticamente)"
 	},
 	"validation": {
 		"apiKey": "Debe proporcionar una clave API válida.",

webview-ui/src/i18n/locales/fr/settings.json

@@ -547,7 +547,8 @@
 	},
 	"thinkingBudget": {
 		"maxTokens": "Tokens maximum",
-		"maxThinkingTokens": "Tokens de réflexion maximum"
+		"maxThinkingTokens": "Tokens de réflexion maximum",
+		"manualThinkingBudget": "Définir un budget manuel de réflexion (sinon le modèle décide automatiquement)"
 	},
 	"validation": {
 		"apiKey": "Vous devez fournir une clé API valide.",

webview-ui/src/i18n/locales/hi/settings.json

@@ -547,7 +547,8 @@
 	},
 	"thinkingBudget": {
 		"maxTokens": "अधिकतम tokens",
-		"maxThinkingTokens": "अधिकतम thinking tokens"
+		"maxThinkingTokens": "अधिकतम thinking tokens",
+		"manualThinkingBudget": "मैन्युअल सोच बजट सेट करें (अन्यथा मॉडल अपने आप निर्णय लेता है)"
 	},
 	"validation": {
 		"apiKey": "आपको एक मान्य API कुंजी प्रदान करनी होगी।",

webview-ui/src/i18n/locales/it/settings.json

@@ -547,7 +547,8 @@
 	},
 	"thinkingBudget": {
 		"maxTokens": "Token massimi",
-		"maxThinkingTokens": "Token massimi di pensiero"
+		"maxThinkingTokens": "Token massimi di pensiero",
+		"manualThinkingBudget": "Imposta budget manuale di pensiero (altrimenti il modello decide automaticamente)"
 	},
 	"validation": {
 		"apiKey": "È necessario fornire una chiave API valida.",

webview-ui/src/i18n/locales/ja/settings.json

@@ -547,7 +547,8 @@
 	},
 	"thinkingBudget": {
 		"maxTokens": "最大 tokens",
-		"maxThinkingTokens": "最大思考 tokens"
+		"maxThinkingTokens": "最大思考 tokens",
+		"manualThinkingBudget": "手動思考予算を設定（そうでなければモデルが自動的に決定）"
 	},
 	"validation": {
 		"apiKey": "有効なAPIキーを入力してください。",

webview-ui/src/i18n/locales/ko/settings.json

@@ -547,7 +547,8 @@
 	},
 	"thinkingBudget": {
 		"maxTokens": "최대 tokens",
-		"maxThinkingTokens": "최대 사고 tokens"
+		"maxThinkingTokens": "최대 사고 tokens",
+		"manualThinkingBudget": "수동 사고 예산 설정 (그렇지 않으면 모델이 자동으로 결정)"
 	},
 	"validation": {
 		"apiKey": "유효한 API 키를 입력해야 합니다.",

webview-ui/src/i18n/locales/nl/settings.json

@@ -547,7 +547,8 @@
 	},
 	"thinkingBudget": {
 		"maxTokens": "Max tokens",
-		"maxThinkingTokens": "Max denk-tokens"
+		"maxThinkingTokens": "Max denk-tokens",
+		"manualThinkingBudget": "Handmatig denkbudget instellen (anders beslist het model automatisch)"
 	},
 	"validation": {
 		"apiKey": "Je moet een geldige API-sleutel opgeven.",

webview-ui/src/i18n/locales/pl/settings.json

@@ -547,7 +547,8 @@
 	},
 	"thinkingBudget": {
 		"maxTokens": "Maksymalna liczba tokenów",
-		"maxThinkingTokens": "Maksymalna liczba tokenów myślenia"
+		"maxThinkingTokens": "Maksymalna liczba tokenów myślenia",
+		"manualThinkingBudget": "Ustaw manualny budżet myślenia (w przeciwnym razie model decyduje automatycznie)"
 	},
 	"validation": {
 		"apiKey": "Musisz podać prawidłowy klucz API.",