Skip to main content

Choose an Assistant Model

Open Settings → Assistant Model to choose the project's default model. The picker groups available models by provider and lets you search by name. The available list depends on the models configured for your installation; it is not limited to one model.

Project defaults​

  • Project model is the default unless a task or individual run selects another model.
  • Reasoning effort offers Low, Medium, and High. Higher effort can take longer and use more credits; it does not guarantee a better result.
  • Generation timeout (minutes) limits how long one generation may run. The default is 60 minutes, with a supported range of 5–120 minutes.

Choose Save to keep your changes. A model marked unavailable needs a different choice or installation configuration before it can be used.

Override a model for one kind of work​

Expand the model overrides to choose a different model for a specific task. Leave a task on Project default to inherit the displayed default. Some lightweight tasks have their own default; the settings show which model will be used.

You can also choose a model when starting generation and in New run → Assistant run → Model. A run override applies to that run and does not change the project default. Playwright execution uses the saved test code and does not have an Assistant model selector.

Hosted Assistant work uses workspace credits. Custom model endpoints are an installation option for self-hosted deployments; see Self-hosted models.

Start with a small, representative test when comparing models. Review the assertions and verification result as well as speed and credits.

See Review generation progress and Credit usage.