Thinking Mode

Let the AI perform deep reasoning before generating its final response, with the option to display the full thinking process.

How to Enable

Click the 💡 lightbulb icon to toggle the global thinking mode on or off.

Three Model States

Toggleable

For example: Qwen's enable_thinking, Gemini's thinkingConfig, Claude's thinking budget, Kimi/Moonshot, MiniMax, etc. — can be toggled via the lightbulb switch.

Always On

For example: DeepSeek Reasoner, OpenAI o1/o3 — thinking is a built-in behavior, and the lightbulb button appears locked.

Unsupported

For example: GPT-4o/GPT-4o-mini, most Ollama/LM Studio models — the lightbulb button is grayed out and unavailable.

Thinking Process Display

Thinking content is shown in a think block that is collapsible and visually separated from the regular response.

Global vs Per-Model

You can use the global lightbulb button for unified control, or override individually in each model's settings via custom JSON. The two are in an OR relationship — use either or both depending on your integration approach.