Coder routes tasks to models based on the agent role and the task. Model selection follows a quality-versus-cost balance rather than a single fixed model for everything.
How routing works
- Each agent role can prefer a model suited to its work (for example, a heavier model for design reasoning, a lighter one for routine edits).
- The router picks a model per task where the capability is supported.
- You can pin or prefer a model for a run where the product allows it.
Cost awareness
Model usage is credit-metered. Routing lighter tasks to lighter models helps control cost while keeping harder tasks on stronger models. The run record reflects the work performed.
Availability
Model routing is part of the Coder execution layer. Specific model availability can change as the product evolves; we avoid stating fixed external model names as if they were permanent. Check the product and status pages for the current supported set.
Related
- Models power the [agent roles](/docs/coder-agents).
- Usage is metered like [cloud runs](/docs/coder-local-cloud-hybrid-runs).
Was this article helpful?