Select a model for a task
- Open the Code composer’s model/effort control.
- Confirm the selected harness. In a new session you can choose another enabled harness; existing-session provider changes have separate restrictions.
- Select a model offered for the current route.
- Choose an available reasoning effort. Unsupported effort levels are not interchangeable across models.
- Review the resulting control before sending the next message.
Reasoning effort changes the model’s requested deliberation level. It is not a provider quota setting, a permission mode, or a guaranteed runtime. Fast mode, where offered, is also separate from permission to edit files.
Set useful defaults
Use Settings → Models for model visibility and custom model configuration, and Code Harnesses for harness defaults. Save those preferences and verify the actual selection in a new task. An existing session can retain its route and other session choices.
The model list comes from the configured route and discovery state. Names shown in marketing examples are not a live catalog for your account. A model available through one API or subscription may be unavailable through another route.
Recover an empty or rejected selection
First distinguish a missing CLI from an unready route and an empty or unverified model catalog. Refresh the relevant harness status if offered. For API models, check the key or deployment under Models. Preserve the error when a model is listed but rejected; do not silently switch to another billing route to make the request run.
See provider setup, API connections, and feature availability.