From ef7abbcb0e95a1a4aa4a970b70195fb04eebc8b9 Mon Sep 17 00:00:00 2001
From: SuYao <sy20010504@gmail.com>
Date: Thu, 1 May 2025 22:57:04 +0800
Subject: [PATCH] Feat/qwen3 support (#5533)

* feat: implement ThinkingPanel for managing reasoning effort and token limits

- Added ThinkingPanel component to handle user settings for reasoning effort and thinking budget.
- Introduced ThinkingSelect and ThinkingSlider components for selecting reasoning effort and adjusting token limits.
- Updated models and hooks to support new reasoning effort and thinking budget features.
- Enhanced Inputbar to integrate ThinkingPanel and provide a toggle for enabling thinking features.
- Updated translations and styles for new components.

* refactor: enhance ThinkingPanel and related components for improved reasoning effort management

- Updated ThinkingPanel to streamline token mapping and error messaging.
- Refactored ThinkingSelect to utilize a list for better UI interaction.
- Enhanced ThinkingSlider with styled components for a more intuitive user experience.
- Adjusted model checks in the configuration to support new reasoning models.
- Improved translations for clarity and consistency across languages.

* feat: close ThinkingPanel on outside click

* feat(icons): add new lightbulb icons and update reasoning effort settings in translations

- Introduced new SVG icons for lightbulb states (off, on at 10%, 50%, and 90%).
- Added "Off" reasoning effort option in English, Japanese, Russian, Simplified Chinese, and Traditional Chinese translations.
- Refactored Inputbar and ThinkingButton components to integrate new reasoning effort logic and icon display.

* feat(thinking): refactor reasoning effort management and introduce new lightbulb icon

- Removed ThinkingPanel, ThinkingSelect, and ThinkingSlider components to streamline reasoning effort management.
- Added new MdiLightbulbAutoOutline icon for the auto reasoning effort option.
- Updated reasoning effort logic in ThinkingButton to support 'auto' and fallback options.
- Enhanced translations for reasoning effort options across multiple languages.
- Adjusted model configurations to integrate new reasoning effort settings.

* refactor(messageThunk): update multi-model dispatch logic for improved assistant handling

- Changed the parameter from baseAssistantId to assistant object for better clarity.
- Updated tasksToQueue structure to include assistant configuration instead of just ID and model.
- Enhanced the creation of assistant messages to utilize the full assistant object.

* chore(messageThunk): remove unused comment for clarity in multi-model response handling

* fix(ThinkingButton): update quick panel title for reasoning effort settings

* fix(GeminiProvider, OpenAIProvider): apply Math.floor to budget calculations for improved accuracy

---------

Co-authored-by: ousugo <dkzyxh@gmail.com>
Co-authored-by: Teo <cheesen.xu@gmail.com>
---
 src/renderer/src/components/Icons/SVGIcon.tsx |  55 ++++++
 src/renderer/src/config/models.ts             | 125 ++++++++++---
 src/renderer/src/i18n/locales/en-us.json      |  19 +-
 src/renderer/src/i18n/locales/ja-jp.json      |  19 +-
 src/renderer/src/i18n/locales/ru-ru.json      |  22 ++-
 src/renderer/src/i18n/locales/zh-cn.json      |  17 +-
 src/renderer/src/i18n/locales/zh-tw.json      |  19 +-
 src/renderer/src/locales/zh/translation.json  |   1 +
 .../src/pages/home/Inputbar/Inputbar.tsx      |  18 +-
 .../pages/home/Inputbar/ThinkingButton.tsx    | 171 ++++++++++++++++++
 .../src/pages/home/Tabs/SettingsTab.tsx       |  94 +---------
 .../AssistantModelSettings.tsx                |  30 +--
 .../providers/AiProvider/AnthropicProvider.ts |  47 +++--
 .../providers/AiProvider/GeminiProvider.ts    |  40 ++--
 .../providers/AiProvider/OpenAIProvider.ts    | 110 ++++++-----
 src/renderer/src/types/index.ts               |  12 +-
 16 files changed, 534 insertions(+), 265 deletions(-)
 create mode 100644 src/renderer/src/locales/zh/translation.json
 create mode 100644 src/renderer/src/pages/home/Inputbar/ThinkingButton.tsx

diff --git a/src/renderer/src/components/Icons/SVGIcon.tsx b/src/renderer/src/components/Icons/SVGIcon.tsx
index e6521a97c7..d58eab7ee5 100644
--- a/src/renderer/src/components/Icons/SVGIcon.tsx
+++ b/src/renderer/src/components/Icons/SVGIcon.tsx
@@ -11,3 +11,58 @@ export const StreamlineGoodHealthAndWellBeing = (props: SVGProps<SVGSVGElement>)
     </svg>
   )
 }
+
+export function MdiLightbulbOffOutline(props: SVGProps<SVGSVGElement>) {
+  return (
+    <svg xmlns="http://www.w3.org/2000/svg" width="1em" height="1em" viewBox="0 0 24 24" {...props}>
+      {/* Icon from Material Design Icons by Pictogrammers - https://github.com/Templarian/MaterialDesign/blob/master/LICENSE */}
+      <path
+        fill="currentColor"
+        d="M12 2C9.76 2 7.78 3.05 6.5 4.68l1.43 1.43C8.84 4.84 10.32 4 12 4a5 5 0 0 1 5 5c0 1.68-.84 3.16-2.11 4.06l1.42 1.44C17.94 13.21 19 11.24 19 9a7 7 0 0 0-7-7M3.28 4L2 5.27L5.04 8.3C5 8.53 5 8.76 5 9c0 2.38 1.19 4.47 3 5.74V17a1 1 0 0 0 1 1h5.73l4 4L20 20.72zm3.95 6.5l5.5 5.5H10v-2.42a5 5 0 0 1-2.77-3.08M9 20v1a1 1 0 0 0 1 1h4a1 1 0 0 0 1-1v-1z"></path>
+    </svg>
+  )
+}
+
+export function MdiLightbulbAutoOutline(props: SVGProps<SVGSVGElement>) {
+  return (
+    <svg xmlns="http://www.w3.org/2000/svg" width="1em" height="1em" viewBox="0 0 24 24" {...props}>
+      {/* Icon from Material Design Icons by Pictogrammers - https://github.com/Templarian/MaterialDesign/blob/master/LICENSE */}
+      <path
+        fill="currentColor"
+        d="M9 2c3.87 0 7 3.13 7 7c0 2.38-1.19 4.47-3 5.74V17c0 .55-.45 1-1 1H6c-.55 0-1-.45-1-1v-2.26C3.19 13.47 2 11.38 2 9c0-3.87 3.13-7 7-7M6 21v-1h6v1c0 .55-.45 1-1 1H7c-.55 0-1-.45-1-1M9 4C6.24 4 4 6.24 4 9c0 2.05 1.23 3.81 3 4.58V16h4v-2.42c1.77-.77 3-2.53 3-4.58c0-2.76-2.24-5-5-5m10 9h-2l-3.2 9h1.9l.7-2h3.2l.7 2h1.9zm-2.15 5.65L18 15l1.15 3.65z"></path>
+    </svg>
+  )
+}
+
+export function MdiLightbulbOn10(props: SVGProps<SVGSVGElement>) {
+  return (
+    <svg xmlns="http://www.w3.org/2000/svg" width="1em" height="1em" viewBox="0 0 24 24" {...props}>
+      {/* Icon from Material Design Icons by Pictogrammers - https://github.com/Templarian/MaterialDesign/blob/master/LICENSE */}
+      <path
+        fill="currentColor"
+        d="M1 11h3v2H1zm18.1-7.5L17 5.6L18.4 7l2.1-2.1zM11 1h2v3h-2zM4.9 3.5L3.5 4.9L5.6 7L7 5.6zM10 22c0 .6.4 1 1 1h2c.6 0 1-.4 1-1v-1h-4zm2-16c-3.3 0-6 2.7-6 6c0 2.2 1.2 4.2 3 5.2V19c0 .6.4 1 1 1h4c.6 0 1-.4 1-1v-1.8c1.8-1 3-3 3-5.2c0-3.3-2.7-6-6-6m1 9.9V17h-2v-1.1c-1.7-.4-3-2-3-3.9c0-2.2 1.8-4 4-4s4 1.8 4 4c0 1.9-1.3 3.4-3 3.9m7-4.9h3v2h-3z"></path>
+    </svg>
+  )
+}
+
+export function MdiLightbulbOn50(props: SVGProps<SVGSVGElement>) {
+  return (
+    <svg xmlns="http://www.w3.org/2000/svg" width="1em" height="1em" viewBox="0 0 24 24" {...props}>
+      {/* Icon from Material Design Icons by Pictogrammers - https://github.com/Templarian/MaterialDesign/blob/master/LICENSE */}
+      <path
+        fill="currentColor"
+        d="M1 11h3v2H1zm9 11c0 .6.4 1 1 1h2c.6 0 1-.4 1-1v-1h-4zm3-21h-2v3h2zM4.9 3.5L3.5 4.9L5.6 7L7 5.6zM20 11v2h3v-2zm-.9-7.5L17 5.6L18.4 7l2.1-2.1zM18 12c0 2.2-1.2 4.2-3 5.2V19c0 .6-.4 1-1 1h-4c-.6 0-1-.4-1-1v-1.8c-1.8-1-3-3-3-5.2c0-3.3 2.7-6 6-6s6 2.7 6 6M8 12c0 .35.05.68.14 1h7.72c.09-.32.14-.65.14-1c0-2.21-1.79-4-4-4s-4 1.79-4 4"></path>
+    </svg>
+  )
+}
+
+export function MdiLightbulbOn90(props: SVGProps<SVGSVGElement>) {
+  return (
+    <svg xmlns="http://www.w3.org/2000/svg" width="1em" height="1em" viewBox="0 0 24 24" {...props}>
+      {/* Icon from Material Design Icons by Pictogrammers - https://github.com/Templarian/MaterialDesign/blob/master/LICENSE */}
+      <path
+        fill="currentColor"
+        d="M7 5.6L5.6 7L3.5 4.9l1.4-1.4zM10 22c0 .6.4 1 1 1h2c.6 0 1-.4 1-1v-1h-4zm-9-9h3v-2H1zM13 1h-2v3h2zm7 10v2h3v-2zm-.9-7.5L17 5.6L18.4 7l2.1-2.1zM18 12c0 2.2-1.2 4.2-3 5.2V19c0 .6-.4 1-1 1h-4c-.6 0-1-.4-1-1v-1.8c-1.8-1-3-3-3-5.2c0-3.3 2.7-6 6-6s6 2.7 6 6m-6-4c-1 0-1.91.38-2.61 1h5.22C13.91 8.38 13 8 12 8"></path>
+    </svg>
+  )
+}
diff --git a/src/renderer/src/config/models.ts b/src/renderer/src/config/models.ts
index c1ed7e84ed..e51a38ab31 100644
--- a/src/renderer/src/config/models.ts
+++ b/src/renderer/src/config/models.ts
@@ -210,6 +210,7 @@ export const FUNCTION_CALLING_MODELS = [
   'o1(?:-[\\w-]+)?',
   'claude',
   'qwen',
+  'qwen3',
   'hunyuan',
   'deepseek',
   'glm-4(?:-[\\w-]+)?',
@@ -2218,30 +2219,40 @@ export function isVisionModel(model: Model): boolean {
   return VISION_REGEX.test(model.id) || model.type?.includes('vision') || false
 }
 
-export function isOpenAIoSeries(model: Model): boolean {
+export function isOpenAIReasoningModel(model: Model): boolean {
   return model.id.includes('o1') || model.id.includes('o3') || model.id.includes('o4')
 }
 
+export function isSupportedReasoningEffortOpenAIModel(model: Model): boolean {
+  return (
+    (model.id.includes('o1') && !(model.id.includes('o1-preview') || model.id.includes('o1-mini'))) ||
+    model.id.includes('o3') ||
+    model.id.includes('o4')
+  )
+}
+
 export function isOpenAIWebSearch(model: Model): boolean {
   return model.id.includes('gpt-4o-search-preview') || model.id.includes('gpt-4o-mini-search-preview')
 }
 
+export function isSupportedThinkingTokenModel(model?: Model): boolean {
+  if (!model) {
+    return false
+  }
+
+  return (
+    isSupportedThinkingTokenGeminiModel(model) ||
+    isSupportedThinkingTokenQwenModel(model) ||
+    isSupportedThinkingTokenClaudeModel(model)
+  )
+}
+
 export function isSupportedReasoningEffortModel(model?: Model): boolean {
   if (!model) {
     return false
   }
 
-  if (
-    model.id.includes('claude-3-7-sonnet') ||
-    model.id.includes('claude-3.7-sonnet') ||
-    isOpenAIoSeries(model) ||
-    isGrokReasoningModel(model) ||
-    isGemini25ReasoningModel(model)
-  ) {
-    return true
-  }
-
-  return false
+  return isSupportedReasoningEffortOpenAIModel(model) || isSupportedReasoningEffortGrokModel(model)
 }
 
 export function isGrokModel(model?: Model): boolean {
@@ -2263,7 +2274,9 @@ export function isGrokReasoningModel(model?: Model): boolean {
   return false
 }
 
-export function isGemini25ReasoningModel(model?: Model): boolean {
+export const isSupportedReasoningEffortGrokModel = isGrokReasoningModel
+
+export function isGeminiReasoningModel(model?: Model): boolean {
   if (!model) {
     return false
   }
@@ -2275,6 +2288,51 @@ export function isGemini25ReasoningModel(model?: Model): boolean {
   return false
 }
 
+export const isSupportedThinkingTokenGeminiModel = isGeminiReasoningModel
+
+export function isQwenReasoningModel(model?: Model): boolean {
+  if (!model) {
+    return false
+  }
+
+  if (isSupportedThinkingTokenQwenModel(model)) {
+    return true
+  }
+
+  if (model.id.includes('qwq') || model.id.includes('qvq')) {
+    return true
+  }
+
+  return false
+}
+
+export function isSupportedThinkingTokenQwenModel(model?: Model): boolean {
+  if (!model) {
+    return false
+  }
+
+  return (
+    model.id.includes('qwen3') ||
+    [
+      'qwen-plus-latest',
+      'qwen-plus-0428',
+      'qwen-plus-2025-04-28',
+      'qwen-turbo-latest',
+      'qwen-turbo-0428',
+      'qwen-turbo-2025-04-28'
+    ].includes(model.id)
+  )
+}
+
+export function isClaudeReasoningModel(model?: Model): boolean {
+  if (!model) {
+    return false
+  }
+  return model.id.includes('claude-3-7-sonnet') || model.id.includes('claude-3.7-sonnet')
+}
+
+export const isSupportedThinkingTokenClaudeModel = isClaudeReasoningModel
+
 export function isReasoningModel(model?: Model): boolean {
   if (!model) {
     return false
@@ -2284,15 +2342,14 @@ export function isReasoningModel(model?: Model): boolean {
     return REASONING_REGEX.test(model.name) || model.type?.includes('reasoning') || false
   }
 
-  if (model.id.includes('claude-3-7-sonnet') || model.id.includes('claude-3.7-sonnet') || isOpenAIoSeries(model)) {
-    return true
-  }
-
-  if (isGemini25ReasoningModel(model)) {
-    return true
-  }
-
-  if (model.id.includes('glm-z1')) {
+  if (
+    isClaudeReasoningModel(model) ||
+    isOpenAIReasoningModel(model) ||
+    isGeminiReasoningModel(model) ||
+    isQwenReasoningModel(model) ||
+    isGrokReasoningModel(model) ||
+    model.id.includes('glm-z1')
+  ) {
     return true
   }
 
@@ -2487,3 +2544,27 @@ export function groupQwenModels(models: Model[]): Record<string, Model[]> {
     {} as Record<string, Model[]>
   )
 }
+
+export const THINKING_TOKEN_MAP: Record<string, { min: number; max: number }> = {
+  // Gemini models
+  'gemini-.*$': { min: 0, max: 24576 },
+
+  // Qwen models
+  'qwen-plus-.*$': { min: 0, max: 38912 },
+  'qwen-turbo-.*$': { min: 0, max: 38912 },
+  'qwen3-0\\.6b$': { min: 0, max: 30720 },
+  'qwen3-1\\.7b$': { min: 0, max: 30720 },
+  'qwen3-.*$': { min: 0, max: 38912 },
+
+  // Claude models
+  'claude-3[.-]7.*sonnet.*$': { min: 0, max: 64000 }
+}
+
+export const findTokenLimit = (modelId: string): { min: number; max: number } | undefined => {
+  for (const [pattern, limits] of Object.entries(THINKING_TOKEN_MAP)) {
+    if (new RegExp(pattern).test(modelId)) {
+      return limits
+    }
+  }
+  return undefined
+}
diff --git a/src/renderer/src/i18n/locales/en-us.json b/src/renderer/src/i18n/locales/en-us.json
index 8877c8de72..f0b33ba941 100644
--- a/src/renderer/src/i18n/locales/en-us.json
+++ b/src/renderer/src/i18n/locales/en-us.json
@@ -56,11 +56,11 @@
       "settings.preset_messages": "Preset Messages",
       "settings.prompt": "Prompt Settings",
       "settings.reasoning_effort": "Reasoning effort",
-      "settings.reasoning_effort.high": "high",
-      "settings.reasoning_effort.low": "low",
-      "settings.reasoning_effort.medium": "medium",
-      "settings.reasoning_effort.off": "off",
-      "settings.reasoning_effort.tip": "Only supported by OpenAI o-series, Anthropic, and Grok reasoning models",
+      "settings.reasoning_effort.off": "Off",
+      "settings.reasoning_effort.high": "Think harder",
+      "settings.reasoning_effort.low": "Think less",
+      "settings.reasoning_effort.medium": "Think normally",
+      "settings.reasoning_effort.default": "Default",
       "settings.more": "Assistant Settings"
     },
     "auth": {
@@ -250,7 +250,14 @@
       "input.upload.upload_from_local": "Upload local file...",
       "input.web_search.builtin": "Model Built-in",
       "input.web_search.builtin.enabled_content": "Use the built-in web search function of the model",
-      "input.web_search.builtin.disabled_content": "The current model does not support web search"
+      "input.web_search.builtin.disabled_content": "The current model does not support web search",
+      "input.thinking": "Thinking",
+      "input.thinking.mode.default": "Default",
+      "input.thinking.mode.default.tip": "The model will automatically determine the number of tokens to think",
+      "input.thinking.mode.custom": "Custom",
+      "input.thinking.mode.custom.tip": "The maximum number of tokens the model can think. Need to consider the context limit of the model, otherwise an error will be reported",
+      "input.thinking.mode.tokens.tip": "Set the number of thinking tokens to use.",
+      "input.thinking.budget_exceeds_max": "Thinking budget exceeds the maximum token number"
     },
     "code_block": {
       "collapse": "Collapse",
diff --git a/src/renderer/src/i18n/locales/ja-jp.json b/src/renderer/src/i18n/locales/ja-jp.json
index 5264cecd30..737856f73a 100644
--- a/src/renderer/src/i18n/locales/ja-jp.json
+++ b/src/renderer/src/i18n/locales/ja-jp.json
@@ -56,11 +56,11 @@
       "settings.preset_messages": "プリセットメッセージ",
       "settings.prompt": "プロンプト設定",
       "settings.reasoning_effort": "思考連鎖の長さ",
-      "settings.reasoning_effort.high": "長い",
-      "settings.reasoning_effort.low": "短い",
-      "settings.reasoning_effort.medium": "中程度",
       "settings.reasoning_effort.off": "オフ",
-      "settings.reasoning_effort.tip": "OpenAI o-series、Anthropic、および Grok の推論モデルのみサポート",
+      "settings.reasoning_effort.high": "最大限の思考",
+      "settings.reasoning_effort.low": "少しの思考",
+      "settings.reasoning_effort.medium": "普通の思考",
+      "settings.reasoning_effort.default": "デフォルト",
       "settings.more": "アシスタント設定"
     },
     "auth": {
@@ -250,7 +250,14 @@
       "input.upload.upload_from_local": "ローカルファイルをアップロード...",
       "input.web_search.builtin": "モデル内蔵",
       "input.web_search.builtin.enabled_content": "モデル内蔵のウェブ検索機能を使用",
-      "input.web_search.builtin.disabled_content": "現在のモデルはウェブ検索をサポートしていません"
+      "input.web_search.builtin.disabled_content": "現在のモデルはウェブ検索をサポートしていません",
+      "input.thinking": "思考",
+      "input.thinking.mode.default": "デフォルト",
+      "input.thinking.mode.custom": "カスタム",
+      "input.thinking.mode.custom.tip": "モデルが最大で思考できるトークン数。モデルのコンテキスト制限を考慮する必要があります。そうしないとエラーが発生します",
+      "input.thinking.mode.default.tip": "モデルが自動的に思考のトークン数を決定します",
+      "input.thinking.mode.tokens.tip": "思考のトークン数を設定します",
+      "input.thinking.budget_exceeds_max": "思考予算が最大トークン数を超えました"
     },
     "code_block": {
       "collapse": "折りたたむ",
@@ -1508,7 +1515,7 @@
         "title": "プライバシー設定",
         "enable_privacy_mode": "匿名エラーレポートとデータ統計の送信"
       },
-      "input.show_translate_confirm": "[to be translated]:显示翻译确认对话框"
+      "input.show_translate_confirm": "翻訳確認ダイアログを表示"
     },
     "translate": {
       "any.language": "任意の言語",
diff --git a/src/renderer/src/i18n/locales/ru-ru.json b/src/renderer/src/i18n/locales/ru-ru.json
index 68022d75ef..7a5e05947f 100644
--- a/src/renderer/src/i18n/locales/ru-ru.json
+++ b/src/renderer/src/i18n/locales/ru-ru.json
@@ -55,12 +55,11 @@
       "settings.model": "Настройки модели",
       "settings.preset_messages": "Предустановленные сообщения",
       "settings.prompt": "Настройки промптов",
-      "settings.reasoning_effort": "Длина цепочки рассуждений",
-      "settings.reasoning_effort.high": "Длинная",
-      "settings.reasoning_effort.low": "Короткая",
-      "settings.reasoning_effort.medium": "Средняя",
-      "settings.reasoning_effort.off": "Выключено",
-      "settings.reasoning_effort.tip": "Поддерживается только моделями рассуждений OpenAI o-series, Anthropic и Grok",
+      "settings.reasoning_effort.off": "Выключить",
+      "settings.reasoning_effort.high": "Стараюсь думать",
+      "settings.reasoning_effort.low": "Меньше думать",
+      "settings.reasoning_effort.medium": "Среднее",
+      "settings.reasoning_effort.default": "По умолчанию",
       "settings.more": "Настройки ассистента"
     },
     "auth": {
@@ -250,7 +249,14 @@
       "input.upload.upload_from_local": "Загрузить локальный файл...",
       "input.web_search.builtin": "Модель встроена",
       "input.web_search.builtin.enabled_content": "Используйте встроенную функцию веб-поиска модели",
-      "input.web_search.builtin.disabled_content": "Текущая модель не поддерживает веб-поиск"
+      "input.web_search.builtin.disabled_content": "Текущая модель не поддерживает веб-поиск",
+      "input.thinking": "Мыслим",
+      "input.thinking.mode.default": "По умолчанию",
+      "input.thinking.mode.default.tip": "Модель автоматически определяет количество токенов для размышления",
+      "input.thinking.mode.custom": "Пользовательский",
+      "input.thinking.mode.custom.tip": "Модель может максимально размышлять количество токенов. Необходимо учитывать ограничение контекста модели, иначе будет ошибка",
+      "input.thinking.mode.tokens.tip": "Установите количество токенов для размышления",
+      "input.thinking.budget_exceeds_max": "Бюджет размышления превышает максимальное количество токенов"
     },
     "code_block": {
       "collapse": "Свернуть",
@@ -1508,7 +1514,7 @@
         "title": "Настройки приватности",
         "enable_privacy_mode": "Анонимная отправка отчетов об ошибках и статистики"
       },
-      "input.show_translate_confirm": "[to be translated]:显示翻译确认对话框"
+      "input.show_translate_confirm": "Показать диалоговое окно подтверждения перевода"
     },
     "translate": {
       "any.language": "Любой язык",
diff --git a/src/renderer/src/i18n/locales/zh-cn.json b/src/renderer/src/i18n/locales/zh-cn.json
index ab91f38448..0e8161cf0e 100644
--- a/src/renderer/src/i18n/locales/zh-cn.json
+++ b/src/renderer/src/i18n/locales/zh-cn.json
@@ -56,11 +56,11 @@
       "settings.preset_messages": "预设消息",
       "settings.prompt": "提示词设置",
       "settings.reasoning_effort": "思维链长度",
-      "settings.reasoning_effort.high": "长",
-      "settings.reasoning_effort.low": "短",
-      "settings.reasoning_effort.medium": "中",
-      "settings.reasoning_effort.off": "关",
-      "settings.reasoning_effort.tip": "仅支持 OpenAI o-series、Anthropic、Grok 推理模型",
+      "settings.reasoning_effort.off": "关闭",
+      "settings.reasoning_effort.low": "浮想",
+      "settings.reasoning_effort.medium": "斟酌",
+      "settings.reasoning_effort.high": "沉思",
+      "settings.reasoning_effort.default": "默认",
       "settings.more": "助手设置"
     },
     "auth": {
@@ -133,6 +133,13 @@
       "input.translating": "翻译中...",
       "input.send": "发送",
       "input.settings": "设置",
+      "input.thinking": "思考",
+      "input.thinking.mode.default": "默认",
+      "input.thinking.mode.default.tip": "模型会自动确定思考的 token 数",
+      "input.thinking.mode.custom": "自定义",
+      "input.thinking.mode.custom.tip": "模型最多可以思考的 token 数。需要考虑模型的上下文限制，否则会报错",
+      "input.thinking.mode.tokens.tip": "设置思考的 token 数",
+      "input.thinking.budget_exceeds_max": "思考预算超过最大 token 数",
       "input.topics": " 话题 ",
       "input.translate": "翻译成{{target_language}}",
       "input.upload": "上传图片或文档",
diff --git a/src/renderer/src/i18n/locales/zh-tw.json b/src/renderer/src/i18n/locales/zh-tw.json
index b2bf98d40d..63515cdd4c 100644
--- a/src/renderer/src/i18n/locales/zh-tw.json
+++ b/src/renderer/src/i18n/locales/zh-tw.json
@@ -56,11 +56,11 @@
       "settings.preset_messages": "預設訊息",
       "settings.prompt": "提示詞設定",
       "settings.reasoning_effort": "思維鏈長度",
-      "settings.reasoning_effort.high": "長",
-      "settings.reasoning_effort.low": "短",
-      "settings.reasoning_effort.medium": "中",
-      "settings.reasoning_effort.off": "關",
-      "settings.reasoning_effort.tip": "僅支援 OpenAI o-series、Anthropic 和 Grok 推理模型",
+      "settings.reasoning_effort.off": "關閉",
+      "settings.reasoning_effort.high": "盡力思考",
+      "settings.reasoning_effort.low": "稍微思考",
+      "settings.reasoning_effort.medium": "正常思考",
+      "settings.reasoning_effort.default": "預設",
       "settings.more": "助手設定"
     },
     "auth": {
@@ -250,7 +250,14 @@
       "input.upload.upload_from_local": "上傳本地文件...",
       "input.web_search.builtin": "模型內置",
       "input.web_search.builtin.enabled_content": "使用模型內置的網路搜尋功能",
-      "input.web_search.builtin.disabled_content": "當前模型不支持網路搜尋功能"
+      "input.web_search.builtin.disabled_content": "當前模型不支持網路搜尋功能",
+      "input.thinking": "思考",
+      "input.thinking.mode.default": "預設",
+      "input.thinking.mode.default.tip": "模型會自動確定思考的 token 數",
+      "input.thinking.mode.custom": "自定義",
+      "input.thinking.mode.custom.tip": "模型最多可以思考的 token 數。需要考慮模型的上下文限制，否則會報錯",
+      "input.thinking.mode.tokens.tip": "設置思考的 token 數",
+      "input.thinking.budget_exceeds_max": "思考預算超過最大 token 數"
     },
     "code_block": {
       "collapse": "折疊",
diff --git a/src/renderer/src/locales/zh/translation.json b/src/renderer/src/locales/zh/translation.json
new file mode 100644
index 0000000000..0519ecba6e
--- /dev/null
+++ b/src/renderer/src/locales/zh/translation.json
@@ -0,0 +1 @@
+ 
\ No newline at end of file
diff --git a/src/renderer/src/pages/home/Inputbar/Inputbar.tsx b/src/renderer/src/pages/home/Inputbar/Inputbar.tsx
index af7aca9faf..da17d02bc1 100644
--- a/src/renderer/src/pages/home/Inputbar/Inputbar.tsx
+++ b/src/renderer/src/pages/home/Inputbar/Inputbar.tsx
@@ -1,7 +1,7 @@
 import { HolderOutlined } from '@ant-design/icons'
 import { QuickPanelListItem, QuickPanelView, useQuickPanel } from '@renderer/components/QuickPanel'
 import TranslateButton from '@renderer/components/TranslateButton'
-import { isGenerateImageModel, isVisionModel, isWebSearchModel } from '@renderer/config/models'
+import { isGenerateImageModel, isReasoningModel, isVisionModel, isWebSearchModel } from '@renderer/config/models'
 import db from '@renderer/databases'
 import { useAssistant } from '@renderer/hooks/useAssistant'
 import { useKnowledgeBases } from '@renderer/hooks/useKnowledge'
@@ -65,6 +65,7 @@ import MentionModelsInput from './MentionModelsInput'
 import NewContextButton from './NewContextButton'
 import QuickPhrasesButton, { QuickPhrasesButtonRef } from './QuickPhrasesButton'
 import SendMessageButton from './SendMessageButton'
+import ThinkingButton, { ThinkingButtonRef } from './ThinkingButton'
 import TokenCount from './TokenCount'
 import WebSearchButton, { WebSearchButtonRef } from './WebSearchButton'
 
@@ -130,7 +131,8 @@ const Inputbar: FC<Props> = ({ assistant: _assistant, setActiveTopic, topic }) =
   const knowledgeBaseButtonRef = useRef<KnowledgeBaseButtonRef>(null)
   const mcpToolsButtonRef = useRef<MCPToolsButtonRef>(null)
   const attachmentButtonRef = useRef<AttachmentButtonRef>(null)
-  const webSearchButtonRef = useRef<WebSearchButtonRef>(null)
+  const webSearchButtonRef = useRef<WebSearchButtonRef | null>(null)
+  const thinkingButtonRef = useRef<ThinkingButtonRef | null>(null)
 
   // eslint-disable-next-line react-hooks/exhaustive-deps
   const debouncedEstimate = useCallback(
@@ -185,6 +187,7 @@ const Inputbar: FC<Props> = ({ assistant: _assistant, setActiveTopic, topic }) =
       const uploadedFiles = await FileManager.uploadFiles(files)
 
       const baseUserMessage: MessageInputBaseParams = { assistant, topic, content: text }
+      console.log('baseUserMessage', baseUserMessage)
 
       // getUserMessage()
       if (uploadedFiles) {
@@ -919,6 +922,14 @@ const Inputbar: FC<Props> = ({ assistant: _assistant, setActiveTopic, topic }) =
                 setFiles={setFiles}
                 ToolbarButton={ToolbarButton}
               />
+              {isReasoningModel(model) && (
+                <ThinkingButton
+                  ref={thinkingButtonRef}
+                  model={model}
+                  assistant={assistant}
+                  ToolbarButton={ToolbarButton}
+                />
+              )}
               <WebSearchButton ref={webSearchButtonRef} assistant={assistant} ToolbarButton={ToolbarButton} />
               {showKnowledgeIcon && (
                 <KnowledgeBaseButton
@@ -1104,7 +1115,8 @@ const ToolbarButton = styled(Button)`
   &.active {
     background-color: var(--color-primary) !important;
     .anticon,
-    .iconfont {
+    .iconfont,
+    .chevron-icon {
       color: var(--color-white-soft);
     }
     &:hover {
diff --git a/src/renderer/src/pages/home/Inputbar/ThinkingButton.tsx b/src/renderer/src/pages/home/Inputbar/ThinkingButton.tsx
new file mode 100644
index 0000000000..623b4a1de2
--- /dev/null
+++ b/src/renderer/src/pages/home/Inputbar/ThinkingButton.tsx
@@ -0,0 +1,171 @@
+import {
+  MdiLightbulbAutoOutline,
+  MdiLightbulbOffOutline,
+  MdiLightbulbOn10,
+  MdiLightbulbOn50,
+  MdiLightbulbOn90
+} from '@renderer/components/Icons/SVGIcon'
+import { useQuickPanel } from '@renderer/components/QuickPanel'
+import { isSupportedReasoningEffortGrokModel, isSupportedThinkingTokenGeminiModel } from '@renderer/config/models'
+import { useAssistant } from '@renderer/hooks/useAssistant'
+import { Assistant, Model, ReasoningEffortOptions } from '@renderer/types'
+import { Tooltip } from 'antd'
+import { FC, ReactElement, useCallback, useEffect, useImperativeHandle, useMemo } from 'react'
+import { useTranslation } from 'react-i18next'
+
+type ThinkingOption = ReasoningEffortOptions | 'off'
+
+export interface ThinkingButtonRef {
+  openQuickPanel: () => void
+}
+
+interface Props {
+  ref?: React.RefObject<ThinkingButtonRef | null>
+  model: Model
+  assistant: Assistant
+  ToolbarButton: any
+}
+
+// 模型类型到支持选项的映射表
+const MODEL_SUPPORTED_OPTIONS: Record<string, ThinkingOption[]> = {
+  default: ['off', 'low', 'medium', 'high'],
+  grok: ['off', 'low', 'high'],
+  gemini: ['off', 'low', 'medium', 'high', 'auto']
+}
+
+// 选项转换映射表：当选项不支持时使用的替代选项
+const OPTION_FALLBACK: Record<ThinkingOption, ThinkingOption> = {
+  off: 'off',
+  low: 'low',
+  medium: 'high', // medium -> high (for Grok models)
+  high: 'high',
+  auto: 'high' // auto -> high (for non-Gemini models)
+}
+
+const ThinkingButton: FC<Props> = ({ ref, model, assistant, ToolbarButton }): ReactElement => {
+  const { t } = useTranslation()
+  const quickPanel = useQuickPanel()
+  const { updateAssistantSettings } = useAssistant(assistant.id)
+
+  const isGrokModel = isSupportedReasoningEffortGrokModel(model)
+  const isGeminiModel = isSupportedThinkingTokenGeminiModel(model)
+
+  const currentReasoningEffort = useMemo(() => {
+    return assistant.settings?.reasoning_effort || 'off'
+  }, [assistant.settings?.reasoning_effort])
+
+  // 确定当前模型支持的选项类型
+  const modelType = useMemo(() => {
+    if (isGeminiModel) return 'gemini'
+    if (isGrokModel) return 'grok'
+    return 'default'
+  }, [isGeminiModel, isGrokModel])
+
+  // 获取当前模型支持的选项
+  const supportedOptions = useMemo(() => {
+    return MODEL_SUPPORTED_OPTIONS[modelType]
+  }, [modelType])
+
+  // 检查当前设置是否与当前模型兼容
+  useEffect(() => {
+    if (currentReasoningEffort && !supportedOptions.includes(currentReasoningEffort)) {
+      // 使用表中定义的替代选项
+      const fallbackOption = OPTION_FALLBACK[currentReasoningEffort as ThinkingOption]
+
+      updateAssistantSettings({
+        reasoning_effort: fallbackOption === 'off' ? undefined : fallbackOption
+      })
+    }
+  }, [currentReasoningEffort, supportedOptions, updateAssistantSettings, model.id])
+
+  const createThinkingIcon = useCallback((option?: ThinkingOption, isActive: boolean = false) => {
+    const iconColor = isActive ? 'var(--color-link)' : 'var(--color-icon)'
+
+    switch (true) {
+      case option === 'low':
+        return <MdiLightbulbOn10 width={18} height={18} style={{ color: iconColor, marginTop: -2 }} />
+      case option === 'medium':
+        return <MdiLightbulbOn50 width={18} height={18} style={{ color: iconColor, marginTop: -2 }} />
+      case option === 'high':
+        return <MdiLightbulbOn90 width={18} height={18} style={{ color: iconColor, marginTop: -2 }} />
+      case option === 'auto':
+        return <MdiLightbulbAutoOutline width={18} height={18} style={{ color: iconColor, marginTop: -2 }} />
+      case option === 'off':
+        return <MdiLightbulbOffOutline width={18} height={18} style={{ color: iconColor, marginTop: -2 }} />
+      default:
+        return <MdiLightbulbOffOutline width={18} height={18} style={{ color: iconColor }} />
+    }
+  }, [])
+
+  const onThinkingChange = useCallback(
+    (option?: ThinkingOption) => {
+      const isEnabled = option !== undefined && option !== 'off'
+      // 然后更新设置
+      if (!isEnabled) {
+        updateAssistantSettings({
+          reasoning_effort: undefined
+        })
+        return
+      }
+      updateAssistantSettings({
+        reasoning_effort: option
+      })
+      return
+    },
+    [updateAssistantSettings]
+  )
+
+  const baseOptions = useMemo(() => {
+    // 使用表中定义的选项创建UI选项
+    return supportedOptions.map((option) => ({
+      level: option,
+      label: t(`assistants.settings.reasoning_effort.${option === 'auto' ? 'default' : option}`),
+      description: '',
+      icon: createThinkingIcon(option),
+      isSelected: currentReasoningEffort === option,
+      action: () => onThinkingChange(option)
+    }))
+  }, [t, createThinkingIcon, currentReasoningEffort, supportedOptions, onThinkingChange])
+
+  const panelItems = baseOptions
+
+  const openQuickPanel = useCallback(() => {
+    quickPanel.open({
+      title: t('assistants.settings.reasoning_effort'),
+      list: panelItems,
+      symbol: 'thinking'
+    })
+  }, [quickPanel, panelItems, t])
+
+  const handleOpenQuickPanel = useCallback(() => {
+    if (quickPanel.isVisible && quickPanel.symbol === 'thinking') {
+      quickPanel.close()
+    } else {
+      openQuickPanel()
+    }
+  }, [openQuickPanel, quickPanel])
+
+  // 获取当前应显示的图标
+  const getThinkingIcon = useCallback(() => {
+    // 如果当前选项不支持，显示回退选项的图标
+    if (currentReasoningEffort && !supportedOptions.includes(currentReasoningEffort)) {
+      const fallbackOption = OPTION_FALLBACK[currentReasoningEffort as ThinkingOption]
+      return createThinkingIcon(fallbackOption, true)
+    }
+    return createThinkingIcon(currentReasoningEffort, currentReasoningEffort !== 'off')
+  }, [createThinkingIcon, currentReasoningEffort, supportedOptions])
+
+  useImperativeHandle(ref, () => ({
+    openQuickPanel
+  }))
+
+  return (
+    <Tooltip placement="top" title={t('assistants.settings.reasoning_effort')} arrow>
+      <ToolbarButton type="text" onClick={handleOpenQuickPanel}>
+        {getThinkingIcon()}
+      </ToolbarButton>
+    </Tooltip>
+  )
+}
+
+export default ThinkingButton
diff --git a/src/renderer/src/pages/home/Tabs/SettingsTab.tsx b/src/renderer/src/pages/home/Tabs/SettingsTab.tsx
index 0bef734c3c..9a21b1c808 100644
--- a/src/renderer/src/pages/home/Tabs/SettingsTab.tsx
+++ b/src/renderer/src/pages/home/Tabs/SettingsTab.tsx
@@ -8,13 +8,11 @@ import {
   isMac,
   isWindows
 } from '@renderer/config/constant'
-import { isGrokReasoningModel, isSupportedReasoningEffortModel } from '@renderer/config/models'
 import { codeThemes } from '@renderer/context/SyntaxHighlighterProvider'
 import { useAssistant } from '@renderer/hooks/useAssistant'
 import { useSettings } from '@renderer/hooks/useSettings'
 import { SettingDivider, SettingRow, SettingRowTitle, SettingSubtitle } from '@renderer/pages/settings'
 import AssistantSettingsPopup from '@renderer/pages/settings/AssistantSettings'
-import { getDefaultModel } from '@renderer/services/AssistantService'
 import { useAppDispatch } from '@renderer/store'
 import {
   SendMessageShortcut,
@@ -52,9 +50,9 @@ import {
   TranslateLanguageVarious
 } from '@renderer/types'
 import { modalConfirm } from '@renderer/utils'
-import { Button, Col, InputNumber, Row, Segmented, Select, Slider, Switch, Tooltip } from 'antd'
+import { Button, Col, InputNumber, Row, Select, Slider, Switch, Tooltip } from 'antd'
 import { CircleHelp, RotateCcw, Settings2 } from 'lucide-react'
-import { FC, useCallback, useEffect, useState } from 'react'
+import { FC, useEffect, useState } from 'react'
 import { useTranslation } from 'react-i18next'
 import styled from 'styled-components'
 
@@ -72,7 +70,6 @@ const SettingsTab: FC<Props> = (props) => {
   const [maxTokens, setMaxTokens] = useState(assistant?.settings?.maxTokens ?? 0)
   const [fontSizeValue, setFontSizeValue] = useState(fontSize)
   const [streamOutput, setStreamOutput] = useState(assistant?.settings?.streamOutput ?? true)
-  const [reasoningEffort, setReasoningEffort] = useState(assistant?.settings?.reasoning_effort)
   const { t } = useTranslation()
 
   const dispatch = useAppDispatch()
@@ -127,17 +124,9 @@ const SettingsTab: FC<Props> = (props) => {
     }
   }
 
-  const onReasoningEffortChange = useCallback(
-    (value?: 'low' | 'medium' | 'high') => {
-      updateAssistantSettings({ reasoning_effort: value })
-    },
-    [updateAssistantSettings]
-  )
-
   const onReset = () => {
     setTemperature(DEFAULT_TEMPERATURE)
     setContextCount(DEFAULT_CONTEXTCOUNT)
-    setReasoningEffort(undefined)
     updateAssistant({
       ...assistant,
       settings: {
@@ -148,7 +137,6 @@ const SettingsTab: FC<Props> = (props) => {
         maxTokens: DEFAULT_MAX_TOKENS,
         streamOutput: true,
         hideMessages: false,
-        reasoning_effort: undefined,
         customParameters: []
       }
     })
@@ -160,25 +148,8 @@ const SettingsTab: FC<Props> = (props) => {
     setEnableMaxTokens(assistant?.settings?.enableMaxTokens ?? false)
     setMaxTokens(assistant?.settings?.maxTokens ?? DEFAULT_MAX_TOKENS)
     setStreamOutput(assistant?.settings?.streamOutput ?? true)
-    setReasoningEffort(assistant?.settings?.reasoning_effort)
   }, [assistant])
 
-  useEffect(() => {
-    // 当是Grok模型时，处理reasoning_effort的设置
-    // For Grok models, only 'low' and 'high' reasoning efforts are supported.
-    // This ensures compatibility with the model's capabilities and avoids unsupported configurations.
-    if (isGrokReasoningModel(assistant?.model || getDefaultModel())) {
-      const currentEffort = assistant?.settings?.reasoning_effort
-      if (!currentEffort || currentEffort === 'low') {
-        setReasoningEffort('low') // Default to 'low' if no effort is set or if it's already 'low'.
-        onReasoningEffortChange('low')
-      } else if (currentEffort === 'medium' || currentEffort === 'high') {
-        setReasoningEffort('high') // Force 'high' for 'medium' or 'high' to simplify the configuration.
-        onReasoningEffortChange('high')
-      }
-    }
-  }, [assistant?.model, assistant?.settings?.reasoning_effort, onReasoningEffortChange])
-
   const formatSliderTooltip = (value?: number) => {
     if (value === undefined) return ''
     return value === 20 ? '∞' : value.toString()
@@ -294,46 +265,6 @@ const SettingsTab: FC<Props> = (props) => {
             </Col>
           </Row>
         )}
-        {isSupportedReasoningEffortModel(assistant?.model || getDefaultModel()) && (
-          <>
-            <SettingDivider />
-            <Row align="middle">
-              <Label>{t('assistants.settings.reasoning_effort')}</Label>
-              <Tooltip title={t('assistants.settings.reasoning_effort.tip')}>
-                <CircleHelp size={14} color="var(--color-text-2)" />
-              </Tooltip>
-            </Row>
-            <Row align="middle" gutter={10}>
-              <Col span={24}>
-                <SegmentedContainer>
-                  <Segmented
-                    value={reasoningEffort || 'off'}
-                    onChange={(value) => {
-                      const typedValue = value === 'off' ? undefined : (value as 'low' | 'medium' | 'high')
-                      setReasoningEffort(typedValue)
-                      onReasoningEffortChange(typedValue)
-                    }}
-                    options={
-                      isGrokReasoningModel(assistant?.model || getDefaultModel())
-                        ? [
-                            { value: 'low', label: t('assistants.settings.reasoning_effort.low') },
-                            { value: 'high', label: t('assistants.settings.reasoning_effort.high') }
-                          ]
-                        : [
-                            { value: 'low', label: t('assistants.settings.reasoning_effort.low') },
-                            { value: 'medium', label: t('assistants.settings.reasoning_effort.medium') },
-                            { value: 'high', label: t('assistants.settings.reasoning_effort.high') },
-                            { value: 'off', label: t('assistants.settings.reasoning_effort.off') }
-                          ]
-                    }
-                    name="group"
-                    block
-                  />
-                </SegmentedContainer>
-              </Col>
-            </Row>
-          </>
-        )}
       </SettingGroup>
       <SettingGroup>
         <SettingSubtitle style={{ marginTop: 0 }}>{t('settings.messages.title')}</SettingSubtitle>
@@ -706,27 +637,6 @@ export const SettingGroup = styled.div<{ theme?: ThemeMode }>`
   margin-bottom: 10px;
 `
 
-// Define the styled component with hover state styling
-const SegmentedContainer = styled.div`
-  margin-top: 5px;
-  .ant-segmented-item {
-    font-size: 12px;
-  }
-  .ant-segmented-item-selected {
-    background-color: var(--color-primary) !important;
-    color: white !important;
-  }
-
-  .ant-segmented-item:hover:not(.ant-segmented-item-selected) {
-    background-color: var(--color-primary-bg) !important;
-    color: var(--color-primary) !important;
-  }
-
-  .ant-segmented-thumb {
-    background-color: var(--color-primary) !important;
-  }
-`
-
 const StyledSelect = styled(Select)`
   .ant-select-selector {
     border-radius: 15px !important;
diff --git a/src/renderer/src/pages/settings/AssistantSettings/AssistantModelSettings.tsx b/src/renderer/src/pages/settings/AssistantSettings/AssistantModelSettings.tsx
index d7604ed79f..24b6025a12 100644
--- a/src/renderer/src/pages/settings/AssistantSettings/AssistantModelSettings.tsx
+++ b/src/renderer/src/pages/settings/AssistantSettings/AssistantModelSettings.tsx
@@ -6,7 +6,7 @@ import { DEFAULT_CONTEXTCOUNT, DEFAULT_TEMPERATURE } from '@renderer/config/cons
 import { SettingRow } from '@renderer/pages/settings'
 import { Assistant, AssistantSettingCustomParameters, AssistantSettings } from '@renderer/types'
 import { modalConfirm } from '@renderer/utils'
-import { Button, Col, Divider, Input, InputNumber, Radio, Row, Select, Slider, Switch, Tooltip } from 'antd'
+import { Button, Col, Divider, Input, InputNumber, Row, Select, Slider, Switch, Tooltip } from 'antd'
 import { isNull } from 'lodash'
 import { FC, useEffect, useRef, useState } from 'react'
 import { useTranslation } from 'react-i18next'
@@ -23,7 +23,6 @@ const AssistantModelSettings: FC<Props> = ({ assistant, updateAssistant, updateA
   const [contextCount, setContextCount] = useState(assistant?.settings?.contextCount ?? DEFAULT_CONTEXTCOUNT)
   const [enableMaxTokens, setEnableMaxTokens] = useState(assistant?.settings?.enableMaxTokens ?? false)
   const [maxTokens, setMaxTokens] = useState(assistant?.settings?.maxTokens ?? 0)
-  const [reasoningEffort, setReasoningEffort] = useState(assistant?.settings?.reasoning_effort)
   const [streamOutput, setStreamOutput] = useState(assistant?.settings?.streamOutput ?? true)
   const [defaultModel, setDefaultModel] = useState(assistant?.defaultModel)
   const [topP, setTopP] = useState(assistant?.settings?.topP ?? 1)
@@ -43,10 +42,6 @@ const AssistantModelSettings: FC<Props> = ({ assistant, updateAssistant, updateA
     }
   }
 
-  const onReasoningEffortChange = (value) => {
-    updateAssistantSettings({ reasoning_effort: value })
-  }
-
   const onContextCountChange = (value) => {
     if (!isNaN(value as number)) {
       updateAssistantSettings({ contextCount: value })
@@ -153,7 +148,6 @@ const AssistantModelSettings: FC<Props> = ({ assistant, updateAssistant, updateA
     setMaxTokens(0)
     setStreamOutput(true)
     setTopP(1)
-    setReasoningEffort(undefined)
     setCustomParameters([])
     updateAssistantSettings({
       temperature: DEFAULT_TEMPERATURE,
@@ -162,7 +156,6 @@ const AssistantModelSettings: FC<Props> = ({ assistant, updateAssistant, updateA
       maxTokens: 0,
       streamOutput: true,
       topP: 1,
-      reasoning_effort: undefined,
       customParameters: []
     })
   }
@@ -383,27 +376,6 @@ const AssistantModelSettings: FC<Props> = ({ assistant, updateAssistant, updateA
         />
       </SettingRow>
       <Divider style={{ margin: '10px 0' }} />
-      <SettingRow style={{ minHeight: 30 }}>
-        <Label>
-          {t('assistants.settings.reasoning_effort')}{' '}
-          <Tooltip title={t('assistants.settings.reasoning_effort.tip')}>
-            <QuestionIcon />
-          </Tooltip>
-        </Label>
-        <Radio.Group
-          value={reasoningEffort}
-          buttonStyle="solid"
-          onChange={(e) => {
-            setReasoningEffort(e.target.value)
-            onReasoningEffortChange(e.target.value)
-          }}>
-          <Radio.Button value="low">{t('assistants.settings.reasoning_effort.low')}</Radio.Button>
-          <Radio.Button value="medium">{t('assistants.settings.reasoning_effort.medium')}</Radio.Button>
-          <Radio.Button value="high">{t('assistants.settings.reasoning_effort.high')}</Radio.Button>
-          <Radio.Button value={undefined}>{t('assistants.settings.reasoning_effort.off')}</Radio.Button>
-        </Radio.Group>
-      </SettingRow>
-      <Divider style={{ margin: '10px 0' }} />
       <SettingRow style={{ minHeight: 30 }}>
         <Label>{t('models.custom_parameters')}</Label>
         <Button icon={<PlusOutlined />} onClick={onAddCustomParameter}>
diff --git a/src/renderer/src/providers/AiProvider/AnthropicProvider.ts b/src/renderer/src/providers/AiProvider/AnthropicProvider.ts
index c2fee1d1e5..07817c953a 100644
--- a/src/renderer/src/providers/AiProvider/AnthropicProvider.ts
+++ b/src/renderer/src/providers/AiProvider/AnthropicProvider.ts
@@ -1,5 +1,5 @@
 import Anthropic from '@anthropic-ai/sdk'
-import { MessageCreateParamsNonStreaming, MessageParam } from '@anthropic-ai/sdk/resources'
+import { MessageCreateParamsNonStreaming, MessageParam, TextBlockParam } from '@anthropic-ai/sdk/resources'
 import { DEFAULT_MAX_TOKENS } from '@renderer/config/constant'
 import { isReasoningModel, isVisionModel } from '@renderer/config/models'
 import { getStoreSetting } from '@renderer/hooks/useSettings'
@@ -10,7 +10,7 @@ import {
   filterEmptyMessages,
   filterUserRoleStartMessages
 } from '@renderer/services/MessagesService'
-import { Assistant, FileTypes, MCPToolResponse, Model, Provider, Suggestion } from '@renderer/types'
+import { Assistant, EFFORT_RATIO, FileTypes, MCPToolResponse, Model, Provider, Suggestion } from '@renderer/types'
 import { ChunkType } from '@renderer/types/chunk'
 import type { Message } from '@renderer/types/newMessage'
 import { removeSpecialCharactersForTopicName } from '@renderer/utils'
@@ -23,8 +23,6 @@ import OpenAI from 'openai'
 import { CompletionsParams } from '.'
 import BaseProvider from './BaseProvider'
 
-type ReasoningEffort = 'high' | 'medium' | 'low'
-
 interface ReasoningConfig {
   type: 'enabled' | 'disabled'
   budget_tokens?: number
@@ -124,32 +122,23 @@ export default class AnthropicProvider extends BaseProvider {
    * @param model - The model
    * @returns The reasoning effort
    */
-  private getReasoningEffort(assistant: Assistant, model: Model): ReasoningConfig | undefined {
+  private getBudgetToken(assistant: Assistant, model: Model): ReasoningConfig | undefined {
     if (!isReasoningModel(model)) {
       return undefined
     }
+    const { maxTokens } = getAssistantSettings(assistant)
 
-    const effortRatios: Record<ReasoningEffort, number> = {
-      high: 0.8,
-      medium: 0.5,
-      low: 0.2
+    const reasoningEffort = assistant?.settings?.reasoning_effort
+
+    if (reasoningEffort === undefined) {
+      return {
+        type: 'disabled'
+      }
     }
 
-    const effort = assistant?.settings?.reasoning_effort as ReasoningEffort
-    const effortRatio = effortRatios[effort]
+    const effortRatio = EFFORT_RATIO[reasoningEffort]
 
-    if (!effortRatio) {
-      return undefined
-    }
-
-    const isClaude37Sonnet = model.id.includes('claude-3-7-sonnet') || model.id.includes('claude-3.7-sonnet')
-
-    if (!isClaude37Sonnet) {
-      return undefined
-    }
-
-    const maxTokens = assistant?.settings?.maxTokens || DEFAULT_MAX_TOKENS
-    const budgetTokens = Math.trunc(Math.max(Math.min(maxTokens * effortRatio, 32000), 1024))
+    const budgetTokens = Math.floor((maxTokens || DEFAULT_MAX_TOKENS) * effortRatio * 0.8)
 
     return {
       type: 'enabled',
@@ -191,6 +180,14 @@ export default class AnthropicProvider extends BaseProvider {
       systemPrompt = buildSystemPrompt(systemPrompt, mcpTools)
     }
 
+    let systemMessage: TextBlockParam | undefined = undefined
+    if (systemPrompt) {
+      systemMessage = {
+        type: 'text',
+        text: systemPrompt
+      }
+    }
+
     const body: MessageCreateParamsNonStreaming = {
       model: model.id,
       messages: userMessages,
@@ -198,9 +195,9 @@ export default class AnthropicProvider extends BaseProvider {
       max_tokens: maxTokens || DEFAULT_MAX_TOKENS,
       temperature: this.getTemperature(assistant, model),
       top_p: this.getTopP(assistant, model),
-      system: systemPrompt,
+      system: systemMessage ? [systemMessage] : undefined,
       // @ts-ignore thinking
-      thinking: this.getReasoningEffort(assistant, model),
+      thinking: this.getBudgetToken(assistant, model),
       ...this.getCustomParameters(assistant)
     }
 
diff --git a/src/renderer/src/providers/AiProvider/GeminiProvider.ts b/src/renderer/src/providers/AiProvider/GeminiProvider.ts
index ef03b27dcc..6ba2f6ad6a 100644
--- a/src/renderer/src/providers/AiProvider/GeminiProvider.ts
+++ b/src/renderer/src/providers/AiProvider/GeminiProvider.ts
@@ -14,7 +14,8 @@ import {
   ToolListUnion
 } from '@google/genai'
 import {
-  isGemini25ReasoningModel,
+  findTokenLimit,
+  isGeminiReasoningModel,
   isGemmaModel,
   isGenerateImageModel,
   isVisionModel,
@@ -31,6 +32,7 @@ import {
 } from '@renderer/services/MessagesService'
 import {
   Assistant,
+  EFFORT_RATIO,
   FileType,
   FileTypes,
   MCPToolResponse,
@@ -54,8 +56,6 @@ import OpenAI from 'openai'
 import { CompletionsParams } from '.'
 import BaseProvider from './BaseProvider'
 
-type ReasoningEffort = 'low' | 'medium' | 'high'
-
 export default class GeminiProvider extends BaseProvider {
   private sdk: GoogleGenAI
 
@@ -216,32 +216,36 @@ export default class GeminiProvider extends BaseProvider {
    * @param model - The model
    * @returns The reasoning effort
    */
-  private getReasoningEffort(assistant: Assistant, model: Model) {
-    if (isGemini25ReasoningModel(model)) {
-      const effortRatios: Record<ReasoningEffort, number> = {
-        high: 1,
-        medium: 0.5,
-        low: 0.2
-      }
-      const effort = assistant?.settings?.reasoning_effort as ReasoningEffort
-      const effortRatio = effortRatios[effort]
-      const maxBudgetToken = 24576 // https://ai.google.dev/gemini-api/docs/thinking
-      const budgetTokens = Math.max(1024, Math.trunc(maxBudgetToken * effortRatio))
-      if (!effortRatio) {
+  private getBudgetToken(assistant: Assistant, model: Model) {
+    if (isGeminiReasoningModel(model)) {
+      const reasoningEffort = assistant?.settings?.reasoning_effort
+
+      // 如果thinking_budget是undefined，不思考
+      if (reasoningEffort === undefined) {
         return {
           thinkingConfig: {
-            thinkingBudget: 0
+            includeThoughts: false
           } as ThinkingConfig
         }
       }
 
+      const effortRatio = EFFORT_RATIO[reasoningEffort]
+
+      if (effortRatio > 1) {
+        return {}
+      }
+
+      const { max } = findTokenLimit(model.id) || { max: 0 }
+
+      // 如果thinking_budget是明确设置的值（包括0），使用该值
       return {
         thinkingConfig: {
-          thinkingBudget: budgetTokens,
+          thinkingBudget: Math.floor(max * effortRatio),
           includeThoughts: true
         } as ThinkingConfig
       }
     }
+
     return {}
   }
 
@@ -313,7 +317,7 @@ export default class GeminiProvider extends BaseProvider {
       topP: assistant?.settings?.topP,
       maxOutputTokens: maxTokens,
       tools: tools,
-      ...this.getReasoningEffort(assistant, model),
+      ...this.getBudgetToken(assistant, model),
       ...this.getCustomParameters(assistant)
     }
 
diff --git a/src/renderer/src/providers/AiProvider/OpenAIProvider.ts b/src/renderer/src/providers/AiProvider/OpenAIProvider.ts
index 8dfd3a9c92..bfb5a75efb 100644
--- a/src/renderer/src/providers/AiProvider/OpenAIProvider.ts
+++ b/src/renderer/src/providers/AiProvider/OpenAIProvider.ts
@@ -1,15 +1,18 @@
-import { DEFAULT_MAX_TOKENS } from '@renderer/config/constant'
 import {
+  findTokenLimit,
   getOpenAIWebSearchParams,
-  isGrokReasoningModel,
   isHunyuanSearchModel,
-  isOpenAIoSeries,
   isOpenAIWebSearch,
   isReasoningModel,
   isSupportedModel,
+  isSupportedReasoningEffortGrokModel,
+  isSupportedReasoningEffortModel,
+  isSupportedReasoningEffortOpenAIModel,
+  isSupportedThinkingTokenClaudeModel,
+  isSupportedThinkingTokenModel,
+  isSupportedThinkingTokenQwenModel,
   isVisionModel,
-  isZhipuModel,
-  OPENAI_NO_SUPPORT_DEV_ROLE_MODELS
+  isZhipuModel
 } from '@renderer/config/models'
 import { getStoreSetting } from '@renderer/hooks/useSettings'
 import i18n from '@renderer/i18n'
@@ -25,6 +28,7 @@ import { processReqMessages } from '@renderer/services/ModelMessageService'
 import store from '@renderer/store'
 import {
   Assistant,
+  EFFORT_RATIO,
   FileTypes,
   GenerateImageParams,
   MCPToolResponse,
@@ -59,8 +63,6 @@ import { FileLike } from 'openai/uploads'
 import { CompletionsParams } from '.'
 import BaseProvider from './BaseProvider'
 
-type ReasoningEffort = 'low' | 'medium' | 'high'
-
 export default class OpenAIProvider extends BaseProvider {
   private sdk: OpenAI
 
@@ -262,8 +264,26 @@ export default class OpenAIProvider extends BaseProvider {
       return {}
     }
 
-    if (isReasoningModel(model)) {
-      if (model.provider === 'openrouter') {
+    if (!isReasoningModel(model)) {
+      return {}
+    }
+    const reasoningEffort = assistant?.settings?.reasoning_effort
+    if (!reasoningEffort) {
+      if (isSupportedThinkingTokenQwenModel(model)) {
+        return { enable_thinking: false }
+      }
+
+      if (isSupportedThinkingTokenClaudeModel(model)) {
+        return { thinking: { type: 'disabled' } }
+      }
+
+      return {}
+    }
+    const effortRatio = EFFORT_RATIO[reasoningEffort]
+    const budgetTokens = Math.floor((findTokenLimit(model.id)?.max || 0) * effortRatio)
+    // OpenRouter models
+    if (model.provider === 'openrouter') {
+      if (isSupportedReasoningEffortModel(model)) {
         return {
           reasoning: {
             effort: assistant?.settings?.reasoning_effort
@@ -271,46 +291,48 @@ export default class OpenAIProvider extends BaseProvider {
         }
       }
 
-      if (isGrokReasoningModel(model)) {
+      if (isSupportedThinkingTokenModel(model)) {
         return {
-          reasoning_effort: assistant?.settings?.reasoning_effort
-        }
-      }
-
-      if (isOpenAIoSeries(model)) {
-        return {
-          reasoning_effort: assistant?.settings?.reasoning_effort
-        }
-      }
-
-      if (model.id.includes('claude-3.7-sonnet') || model.id.includes('claude-3-7-sonnet')) {
-        const effortRatios: Record<ReasoningEffort, number> = {
-          high: 0.8,
-          medium: 0.5,
-          low: 0.2
-        }
-
-        const effort = assistant?.settings?.reasoning_effort as ReasoningEffort
-        const effortRatio = effortRatios[effort]
-
-        if (!effortRatio) {
-          return {}
-        }
-
-        const maxTokens = assistant?.settings?.maxTokens || DEFAULT_MAX_TOKENS
-        const budgetTokens = Math.trunc(Math.max(Math.min(maxTokens * effortRatio, 32000), 1024))
-
-        return {
-          thinking: {
-            type: 'enabled',
-            budget_tokens: budgetTokens
+          reasoning: {
+            max_tokens: budgetTokens
           }
         }
       }
-
-      return {}
     }
 
+    // Qwen models
+    if (isSupportedThinkingTokenQwenModel(model)) {
+      return {
+        enable_thinking: true,
+        thinking_budget: budgetTokens
+      }
+    }
+
+    // Grok models
+    if (isSupportedReasoningEffortGrokModel(model)) {
+      return {
+        reasoning_effort: assistant?.settings?.reasoning_effort
+      }
+    }
+
+    // OpenAI models
+    if (isSupportedReasoningEffortOpenAIModel(model)) {
+      return {
+        reasoning_effort: assistant?.settings?.reasoning_effort
+      }
+    }
+
+    // Claude models
+    if (isSupportedThinkingTokenClaudeModel(model)) {
+      return {
+        thinking: {
+          type: 'enabled',
+          budget_tokens: budgetTokens
+        }
+      }
+    }
+
+    // Default case: no special thinking settings
     return {}
   }
 
@@ -343,7 +365,7 @@ export default class OpenAIProvider extends BaseProvider {
     const isEnabledWebSearch = assistant.enableWebSearch || !!assistant.webSearchProviderId
     messages = addImageFileToContents(messages)
     let systemMessage = { role: 'system', content: assistant.prompt || '' }
-    if (isOpenAIoSeries(model) && !OPENAI_NO_SUPPORT_DEV_ROLE_MODELS.includes(model.id)) {
+    if (isSupportedReasoningEffortOpenAIModel(model)) {
       systemMessage = {
         role: 'developer',
         content: `Formatting re-enabled${systemMessage ? '\n' + systemMessage.content : ''}`
diff --git a/src/renderer/src/types/index.ts b/src/renderer/src/types/index.ts
index 889333f49a..6d889756a9 100644
--- a/src/renderer/src/types/index.ts
+++ b/src/renderer/src/types/index.ts
@@ -36,6 +36,16 @@ export type AssistantSettingCustomParameters = {
   type: 'string' | 'number' | 'boolean' | 'json'
 }
 
+export type ReasoningEffortOptions = 'low' | 'medium' | 'high' | 'auto'
+export type EffortRatio = Record<ReasoningEffortOptions, number>
+
+export const EFFORT_RATIO: EffortRatio = {
+  low: 0.2,
+  medium: 0.5,
+  high: 1,
+  auto: 2
+}
+
 export type AssistantSettings = {
   contextCount: number
   temperature: number
@@ -46,7 +56,7 @@ export type AssistantSettings = {
   hideMessages: boolean
   defaultModel?: Model
   customParameters?: AssistantSettingCustomParameters[]
-  reasoning_effort?: 'low' | 'medium' | 'high'
+  reasoning_effort?: ReasoningEffortOptions
 }
 
 export type Agent = Omit<Assistant, 'model'> & {