Auto starts answering sooner with lighter reasoning

Auto now asks its models to reason less before they start working. Balanced and Max each drop one reasoning level, and so does every tier when Auto runs on a subscription you connected. Medium is the lowest level Auto ever uses, so Fast keeps its models at medium. The tiers still choose the same models; they spend less time thinking silently first, so replies and each step of a workflow begin sooner. The change is largest on GPT-6.1 Sol: in our tests, moving Max from extra high to high reasoning cut a multi-step reasoning task from about 21 seconds to about 12. Read how Auto chooses models on the models page.

How it works

• Fast is unchanged: Claude Haiku 5.5, GPT-6 Luna, Gemini 3.8 Flash, and Grok 4.7 at medium.
• Balanced runs Claude Sonnet 5.5, GPT-6.1 Sol, Grok 4.7, and Gemini 3.8 Flash at medium, down from high.
• Max runs Claude Opus 5.5, Grok 4.7, and Gemini 3.8 Flash at medium, and GPT-6.1 Sol at high, down from extra high.
• Your own subscriptions drop a level too. When Auto runs on a Claude, ChatGPT, Grok, or Gemini subscription you connected through Token Tunnel, Fast and Balanced now reason at medium and Max at high (Gemini at medium), each one level lower than before.
• Picking a model directly is unchanged. It keeps its own default, and the reasoning picker beside the model still sets any level for a conversation or workflow that needs more depth.