Two changes to the front of every task: how much thinking it gets, and how clearly it starts.

Choosing how much reasoning a task deserves

Not every request needs the same depth. Model reasoning controls let you set how much reasoning goes into a given task, and intelligence settings now work inside group chats too, so a group can pick the model capability that suits the work it is actually doing together.

Turning a rough thought into a clear instruction

Most people start with a rough idea rather than a structured brief, and then get a result shaped by what they left out.

Enhance Prompt closes that gap before the agent begins. The upgraded prompt-enhancement flow takes your initial thought and gives the agent a better read on the outcome you want, the context that matters, and the level of detail you expect. You get strong results without needing to be an expert prompter.

A clearer view of what the agent is doing

We heard the feedback on reasoning. Raw multi-step output is useful, and it can also be too dense to follow.

The agent reasoning timeline is now compressed into a more structured view. In progress, completed and error states are easier to tell apart as an agent works, and the steps that actually help you understand what is happening are prioritised over every low-level intermediate action shown at equal weight.

Intelligence Controls and Enhance Prompt: Better Input, Better Work