ThinkingLevelSupportEN

Overview

• Some models support both thinking mode and normal mode
• You can switch by setting the Thinking Level in the API Settings
• Setting different thinking levels (OFF/LOW/MEDIUM/HIGH) will affect the model's thinking behavior, response time, and Token consumption

API Settings

In Model Management -> Basic Settings, you can find the Thinking Level configuration item

• OFF: Disable the thinking feature and use the standard response mode
• LOW: Enable thinking, allocate a smaller Token budget (e.g., 1024 Tokens), suitable for simple logic or saving consumption
• MEDIUM: Allocate a medium Token budget (e.g., 1536 Tokens), balancing translation quality and speed
• HIGH: Allocate the maximum Token budget (e.g., 2048 Tokens or higher), suitable for complex semantic understanding and rare topics

Supported Models

• Models that only support the thinking toggle
  • GLM series
    • LOW MEDIUM HIGH - ON
    • OFF - OFF
  • Kimi series
    • LOW MEDIUM HIGH - ON
    • OFF - OFF
  • DeepSeek series
    • LOW MEDIUM HIGH - ON
    • OFF - OFF
• Models that only support thinking levels
  • Gemini 2.5 Pro / Gemini 3 Pro / Gemini 3.1 Pro
    • HIGH - Takes effect normally
    • OFF LOW MEDIUM - Equivalent to LOW
• Models that fully support both thinking toggle and thinking levels
  • doubao-seed-1.6 / doubao-seed-1.8 / doubao-seed-2.0 series
  • Gemini 2.5 Flash / Gemini 3 Flash
  • Claude Sonnet 3.7 / Claude Haiku 4.x / Claude Sonnet 4.x / Claude Opus 4.x