Composer Predictions in Codex

Composer predictions is a beta feature in OpenAI Codex that suggests a user's next message based on the conversation.^[1]

Availability

Chat on web or desktop is excluded.^[2]

Using predictions

After Codex replies, a prediction may appear in the message box. Tab inserts it without sending. Users can edit it, submit it, or write something else. Predictions propose whole messages, not word-by-word completion, and never automatically send follow-ups.^[2]

The default-on switch is Show predictions in Settings > General > Composer.^[2]

Context and usage

Predictions use the current thread, including memory or connected-app context already present. They do not retrieve those sources independently and work with memory disabled.^[2]

During the beta, generating predictions costs nothing extra and consumes neither credits nor Codex usage allowance. Sending accepted text incurs normal usage limits and billing.^[2]

Missing predictions

Suggestions may take seconds, especially in long threads, or be skipped. Check the app version, setting, account, model, and thread type.^[2]

References

1. ^OpenAI Developers. Composer predictions beta announcement. X.
2. ^1 ^2 ^3 ^4 ^5 ^6OpenAI Help Center. Composer predictions in Codex. Accessed October 12, 2026.