How Auto picks a model
Auto picks a model for each message, so you don’t have to know which model is best at what.
What Auto looks at
Auto first works out what kind of request your message is — for example an image to create, an attached image or document, a coding task, maths or data analysis, creative writing, a translation, a summary, a long paste, or a quick everyday question.
For each kind of request there is a set of models that handle it well. Auto sends your message to the lowest-priced model in that set that is available to you right now and that your balance can cover.
When your weekly premium credit is nearly used up, Auto leans towards lower-priced models for everyday requests (the reply says “saving credits”). It doesn’t do this for images, code, maths, data or documents.
Auto for one company
Choose Auto · Claude, Auto · ChatGPT and so on to stay with one company. Auto then picks that company’s best-suited model for the request, and its lowest-priced one for quick requests such as short questions, translations and summaries.
If that company can’t serve the request — it doesn’t make images, isn’t available in your region, or is down — Auto picks another company’s model and tells you, for example: “Claude doesn’t make images — Auto picked GPT Image 1.5”.
You always see what Auto did
- While you type, a line under the message box previews the choice: “Auto → {model} ({reason})”.
- Every reply’s label says which model answered and why, for example “Auto → {model} · coding task”.
What it costs
An Auto reply costs exactly what the model that answered costs. Auto adds no fee of its own.
Turning Auto off
Pick a specific model in the picker. That model is then used for this chat until you change it. To make Auto (or a specific model) the default for new chats, use Settings → Models.