Chat
Models and thinking levels
Choose between Sonnet 5, Opus 4.8, Fable 5, and GPT 5.5, and set how deeply the model thinks per task.
The model picker sits in the chat composer, on the dashboard and inside every project. It selects which AI model powers the agent, and, inside the same card, how much the model thinks before acting. Each model shows its speed, intelligence, and cost profile so you can pick the right tool per task.
Available models
| Model | Best for | Availability |
|---|---|---|
| Sonnet 5 | Fast, efficient building: everyday edits, iterations, and most features | All plans |
| Opus 4.8 | Deeper reasoning: complex features, multi-file refactors, tricky debugging | All plans |
| Fable 5 | The most capable model for the hardest work: architecture, subtle bugs, polish | Pro plans |
| GPT 5.5 | OpenAI's flagship, if you prefer its style of reasoning and output | All plans |
Bigger models and deeper thinking consume more credits per message. A good default: Sonnet 5 for iteration speed, stepping up when a task resists.
Thinking levels
Thinking levels control how much reasoning the model does before answering. They live as chips inside the active model's row in the picker, and the picker button shows your current level.
| Level | What it means |
|---|---|
| Auto | The model decides how much to think based on the task. The default, right for most work. |
| Extended | Maximum reasoning for the hardest problems. Slower and costs more credits. |
| Balanced | A fixed middle ground between speed and depth. |
| Quick | Minimal thinking for simple changes: copy edits, small tweaks, fast iterations. |
Where your choice applies
- Your model and thinking level persist per project; switching projects keeps each project's own choice.
- Picking a model on the homepage or dashboard composer carries into the project it creates.
- On the Free plan, Pro-only models are shown but locked; everything else is available.