Auto picks the right model for the job

Until now, choosing Auto as the model always ran your thread on one fixed default. Now Auto reads your request first: a quick routing step picks the most suitable of the available models — a stronger model for complex reasoning or coding work, a faster one for short and simple asks — and the round runs on that choice.

The routing is careful by design: if it cannot decide quickly, your round simply runs on the familiar default, so Auto is never slower or less reliable than before. The chosen model is recorded on the response, so you can always see which model served a round.