Published: October 13, 2026
What happened
Composer predictions in Codex suggests a possible next prompt after a response finishes. The suggestion uses the current thread context and appears in the message box; the user can accept it with Tab, edit it, or ignore it.
How it works
The feature does not send text automatically or continue a task without a human decision. That matters in long engineering sessions: a suggestion can preserve direction, while the final message remains an action the user can review.
Context
In beta, it is available to adults on personal Pro plans in local and SSH threads using GPT-6 Astra or GPT-6.1 Sol. OpenAI says generating predictions does not consume Codex limits, while messages that are sent follow normal limits.
Why it matters
The suggestion remains a proposal rather than an automatic action and should not silently expand the task.