Skip to main content

Model Orchestrator uses less of your budget for things you use

Interactive requests go to GPT-6.2 Sol, everyday and demanding ones alike. Auto in the model picker is now Orchestrator, and replies stop "correcting" an earlier model's name.

Written by Babbily Support

Requests for something to use rather than read, such as a calculator, a quiz, a checklist, a picker or a simulator, go to GPT-6.2 Sol at Medium reasoning, everyday and demanding ones alike. On everyday ones Gemini 4 Neon did about as well but used about 4 times as much of your API usage budget, and on demanding ones Sol did better than Neon at about a fifth of the cost. Simple requests still start on Claude Haiku 5.5.

From now on, when two models do about as well on a kind of work, Orchestrator uses the one that uses less of your budget. Learn how Model Orchestrator works.

Auto is now Orchestrator

The choice at the top of the model picker that picks a model for each message is now called Orchestrator, so it's easier to tell apart from Auto tool mode, which picks tools. It works the same way, and your chats and settings carry over. Tool mode keeps the name Auto. See how Model Orchestrator works.

Fixes

  • No more "corrections" when Model Orchestrator switches models. When Orchestrator picked a different model partway through a chat, a reply could end by "correcting" the model named in an earlier reply. Replies now know Orchestrator can pick a different model for each message, and mention which model answered only when you ask.

Did this answer your question?