Thinking effort
Choose how much a Claude model thinks before it answers, from Low to Max, and when Ultracode is available.
Thinking effort sets how deeply the model reasons before it answers. More effort gives better results on hard problems, but turns take longer.
Where effort is available
The Thinking effort control appears only for models that support it:
| Model | Effort levels |
|---|---|
| Claude Opus 5.5, Claude Sonnet 5.5, Claude Fable 5.1 | Low, Medium, High, Extra High, Max |
| Claude Opus 5.5 with a Claude Pro/Max sign-in | The levels above, plus Ultracode |
| Claude Haiku 4.5 | No effort control |
| Lyna models, OpenAI models, Kimi K3 | No effort control |
The levels
| Level | What it does |
|---|---|
| Low | Rarely reasons. Fastest answers. |
| Medium | Reasons on harder problems only. |
| High | Reasons thoroughly on most problems. This is the default and the recommended level. |
| Extra High | Deeper reasoning, just below Max. |
| Max | Always reasons. |
| Ultracode | Max effort plus agent teams, for this session only. |
Change the effort
- Open Thinking effort in the composer and pick a level.
- Or type
/effortfollowed by a level:low,medium,high,xhigh,maxorultracode.
The chat confirms the new level. If the current model has no effort control, the chat tells you.
Ultracode
Ultracode turns on agent teams for the current session, so the AI can split a large task across parallel sub-agents. It resets when the session ends. See Agent teams.
Tips
- Keep High for most work.
- Drop to Low or Medium for quick copy or style edits.
- Use Extra High or Max for tricky bugs and architecture changes.
Next steps
Models and providers
Pick the AI model for your project. Use the built-in Lyna models, or connect your own Claude, ChatGPT or Kimi Code account or API key.
Attachments and screenshots
Show the AI what you mean. Attach images, capture the preview, mark it up, import Figma frames or dictate your message.