How Auto picks a model, and when to pin one instead

Why Auto exists

Frontier models are brilliant at hard problems and wasteful on easy ones. Renaming a variable or writing a docstring doesn't need the most capable model on the market, and using one makes the request slower and more expensive. Auto exists so you don't have to make that call dozens of times a day.

What Auto looks at

For every request, Auto weighs a few things:

• The kind of task. An inline edit, a question in chat and a multi-file agent run have very different needs.
• How much context it needs. Requests that span a large part of the codebase go to models with large context windows.
• How hard it looks. Signals like failing tests, long stack traces or a request to redesign something push towards stronger reasoning models.
• Your settings. Pinned models, team allowlists and your spend cap always come first.

Why it doesn't always pick the biggest model

Because bigger isn't always better. For small, well-defined edits, fast models are often just as accurate and answer in a fraction of the time. Since on-demand usage is billed at list price, choosing the cheaper model also stretches your included usage further.

If a fast model struggles with something, ask again with a stronger model from the picker. The conversation keeps its context, so nothing is lost.

When to pin a model instead

• You have a favourite for a specific job. Many people pin Claude Sonnet for agent work and let Auto handle everything else.
• You need consistent behaviour. For long refactors, staying on one model keeps the style of the changes consistent.
• You're comparing models. Pin one, then switch mid-conversation to see how another handles the same context.
• Your team requires it. On Teams and Enterprise, admins can limit which models are available.

Seeing what Auto picked

Every response is labelled with the model that produced it, along with its token count and cost, for example "Claude Sonnet · picked by Auto · $0.07". The usage dashboard breaks spending down by model, so you can see exactly how Auto spends your included usage. Browse every available model.