Codex tests composer predictions: it drafts your next message after each reply, Tab to accept; personal Pro only, two flagship models, free of usage limits in beta

Why it matters

【Key points】

At 02:22 Beijing time on Oct 10, OpenAI Developers announced a beta of composer predictions in the Codex desktop app. After Codex responds, it may pre-fill a suggested next message based on the conversation and how you talk to it; press Tab to accept, edit, or ignore it.

Per OpenAI's help article, the beta is limited to adults on personal ChatGPT Pro plans, in local and SSH threads in the Codex desktop app, with GPT-6 Astra or GPT-6.1 Sol. It is on by default and can be turned off under Settings, General, Composer.

During the beta, generating predictions costs nothing extra and does not count toward Codex usage limits; sent messages are billed normally. Predictions use only the current thread, do not look up memories or apps on their own, and are never sent automatically.

【Verifiable】

OpenAI Developers post: https://x.com/OpenAIDevs/status/2108624138369929725

OpenAI help article: https://help.openai.com/en/articles/20001601

~80 posts and 81 quotes in the first half hour; reactions split between high acceptance rates and complaints about Pro-only access.

【Our read】

1. Autocomplete is moving from code to the conversation. Tab used to finish the next line of code; now it drafts the user's next instruction to the agent, a sign that many turns in long coding sessions are predictable: "continue", "run the tests", "fix that error".

2. The human in the loop is thinning. OpenAI deliberately keeps a manual send and no auto-continue; that line matters. If predictions were ever sent automatically, Codex would be talking to itself and review would shift entirely to after the fact.

3. Pro-only, two flagship models and no usage charge in beta suggest each prediction is a costly extra inference. Watch whether it reaches Plus and enterprise plans, and whether acceptance data is used to train models to anticipate user intent.