Codex introduced composer predictions in beta on October 9. After the assistant replies, a suggested next message can appear in the composer. Pressing Tab accepts the suggestion into the message box; it does not send it.
OpenAI's feature guidance describes predictions as a way to continue the current conversation. The dated release notes announced the beta for personal ChatGPT Pro users aged 18 and over, using the latest Codex desktop app and supported models.
Sending makes the suggested text the next instruction, so it needs the same review as a message the user writes.
For example, a conversation about checking a change can continue toward another review, a local fix, or a release. Those are different instructions. Read the verb in a proposed follow-up before sending it, particularly when the next step would change a shared system.
The help page specifies local and SSH chats in the desktop app with GPT-6 Astra and GPT-6.1 Sol. It also says the composer can skip a prediction. An empty suggestion area therefore does not establish that the feature is broken or that the account lacks access. The beta's age, plan, client and model conditions should be checked together before comparing two users' screens.
Predictions can be turned off through Show predictions in the Composer section of General settings. OpenAI says generating them does not consume the beta user's Codex allowance or credits; messages that are actually sent remain subject to normal usage rules.
What the next action does
| Action | Result |
|---|---|
| Press Tab | Put the prediction in the composer |
| Edit the draft | Change the instruction before sending |
| Send the message | Give the assistant the actual next instruction |
If a prediction assumes a release or an external change, replace that assumption with the step you intend to authorize.
