Models and Auto routing
Connect accounts in Settings → Providers, then choose a model in the session composer. Auto routing can adapt the model or reasoning effort for each message.
Choose a routing strategy
Open Settings → Auto routing.
| Option | What changes |
|---|---|
| No auto routing | Keeps your chosen provider, model, and reasoning. |
| Reasoning only | Keeps one model for cache reuse and adapts its reasoning level. |
| Model and reasoning | Chooses a model and reasoning within the selected provider. Experimental. |
| Custom routing | Uses the provider, model, and reasoning you configure for each task difficulty. |
Try reasoning-only routing
- Connect a provider account and make sure its models are available.
- Select Reasoning only under Auto routing.
- Choose the provider account and model.
- Select Auto routing in the composer and send a small task.
- Check the routing indicator, then try a task that needs more analysis.
Customize difficulty routes
Select Custom routing and expand each difficulty. “Follow provider order” uses the first available account. “Account default” follows the account's current default model; a pinned model stays on that exact model.
Classify ambiguous messages
With this option enabled, a quick classifier helps when local cues are inconclusive. When disabled, routing uses local heuristics and ambiguous messages use Normal.
Cache and cost
Changing models between messages can reduce prompt-cache reuse and increase cost or latency. Use Reasoning only when keeping the same model is important.
If Auto is missing
Check that routing is enabled and a provider account is available. With No auto routing selected, choose a concrete model instead.