Codex waiting on your next message? Tab to accept composer predictions (Pro beta)

#What changed

Once Codex finishes a response, the composer can now pre-fill a full suggested next message, drafted from what is already in the current thread. Press Tab to accept it. Accepting only drops the text into the composer: nothing is sent until you review it, edit it if needed, and send it yourself.

Source: OpenAI Developers (@OpenAIDevs), checked against the OpenAI Help Center article Composer predictions in Codex.

#Who is affected

Personal ChatGPT Pro users (18+) using the Codex desktop app, in local and SSH threads running GPT-6 Astra or GPT-6.1 Sol. The feature is on by default for eligible users.

Predictions do not appear in Chat on web or Desktop. Work may show them, but only in eligible local Work threads.

#What to do now

• When a prediction fits, press Tab, adjust the text, and send.
• Usage: during the beta, generating predictions does not count against your Codex usage limits. Sending a message, accepted or not, still uses normal usage and billing.
• To turn it off, go to Settings → General → Composer → Show predictions.
• Privacy: predictions draw on the thread's context only. They do not pull in memories or connected apps on their own.

#What is not confirmed

How long the beta lasts, whether predictions will count toward usage after it ends, and whether other plans or models will get the feature.