How the Auto-Router Picks Your Model
Auto mode answers the question most people do not want to think about per message: which of 200+ models should take this one? Here is what it does, so you can trust it where it earns trust and pin a model where you know better.
What Auto does
When a chat is set to Auto, the router classifies the message and routes it to a model tier: routine prompts go to fast, affordable models; hard prompts go to stronger ones. Two flavors exist:
- Auto Smart: the everyday default, biased toward speed and economy.
- Auto Expert: biased toward capability for consistently hard work.
The reply shows which model answered, always: routing is never a secret.
What it optimizes
The router's job is fit, not favoritism: match the prompt's difficulty to the cheapest tier that handles it well. Simple lookups on a frontier model waste money; hard reasoning on a budget model wastes your time. The pre-send cost estimate reflects the routed model, so economics stay visible.
When to pin instead
Auto is a default, not a doctrine. Pin a specific model when:
- Consistency matters: long-form writing mid-piece should not change voice.
- You know the winner: your side-by-side testing crowned a model for this job.
- The task is niche: a coding specialist for a long coding session beats re-classification per message.
Switching is per-message: pinning now does not pin forever, and context follows the switch.
Good to know
- Agents can be set to Auto or pinned per agent; pinning suits specialists.
- Provider preferences and blocks apply to routed choices like everything else.
- Auto never routes to your local models implicitly; local routing is its own explicit setting (see local inference).
Run it for a real day: leave a general chat on Auto, watch which models answer, and pin only where you disagree with its calls. The choosing models help article covers the full mechanics.
Found this helpful? Share it: