How to Use Codex Composer Predictions to Accept Your Next Prompt with Tab
Codex Composer Predictions can suggest a draft for your next message in the composer after Codex replies. Press Tab to make the suggestion editable, then review it and send it yourself.
What is Composer Predictions?
The feature proposes a complete next message in an existing conversation. Starter suggestions on a new chat help you choose a first request, while word-by-word autocomplete extends text you are already typing. Information from earlier work or connected apps can inform a suggestion when it is already present in the thread. Predictions also work with memory turned off. (OpenAI’s guide to using Composer Predictions)
How to accept and edit a suggestion with Tab
- Send a message in a supported thread and look in the composer once Codex finishes responding.
- If the suggestion fits your next step, press Tab to insert it as an editable draft.
- Check who or what the request refers to, its scope, and the output format. Edit as needed, then send it yourself. You can start typing your own message at any time.
When is it useful? A blog writing example
Suppose you have just received an outline for a comparison of GPT-6.2 Astra and Sol and see the suggestion “Put the costs of the two models in a table.” Accept it with Tab and add “Separate API token rates from cost per task, and give the measurement conditions for the latter.” When the table is ready, check that it separates standard API prices from costs observed in a particular evaluation.
Who can use it, and why might a prediction be missing?
OpenAI’s guidance checked on November 18, 2026 lists a beta for adults aged 18 or older on personal ChatGPT Pro accounts in local or SSH threads in Codex desktop. Supported models include GPT-6.2 Astra and Sol, GPT-6 Astra, GPT-6.1 Sol, and GPT-6.1 Terra. Regular Chat on web and desktop is outside this rollout. Predictions may appear in local Work conversations. (OpenAI’s availability and settings guide)
Update the app and check that Show predictions is enabled in Settings → General → Composer. Then check your account, model, and thread type. A suggestion may take several seconds, or longer in a long conversation, and will not appear after every response. You can turn the feature off and back on in the same place.




