ERA OF MINDS

Codex composer predictions: how OpenAI's next-message suggestions work

By Era of Minds ·

Short answer

Codex composer predictions is a beta feature in the Codex desktop app that suggests your next message after Codex finishes a response. It reads the conversation in your current thread and places a suggestion in the message box. Press Tab to accept, then edit or send. Nothing sends on its own. It's for personal ChatGPT Pro users.

What are Codex composer predictions?

When Codex finishes replying, a suggested next message can appear in the box where you type. You can take it, change it, or ignore it and write your own. OpenAI's help article calls it a feature that "suggests what you might want to say next, based on the conversation in your current Codex thread."

OpenAI Developers announced it on October 9, 2026, saying Codex "can now suggest your next message based on your conversation and how you talk to it." The team added that it was "one of the most loved new features we've ever tested internally."

A prediction won't show up after every reply. The help article says the composer may choose to skip one, and that this alone doesn't mean anything is broken.

Who can use Codex composer predictions?

The beta is narrow. You need all of these:

  • A personal ChatGPT Pro account, and you must be 18 or older.
  • The Codex desktop app, updated to the latest version.
  • A local or SSH thread. The follow-up post describes it as "for local tasks."
  • A supported model: GPT-6 Astra or GPT-6.1 Sol.

It's available in all supported regions. It does not appear in Chat on the web or desktop, and it isn't offered on other plans during the beta.

How do you accept or turn off Codex predictions?

Press Tab. The suggestion drops into the message box, where you can read it, edit it and send it when ready. Accepting only adds text. It never sends the message for you, and predictions can't continue a task on their own.

You don't have to wait for one, either. Start typing and your own message takes over.

Predictions are on by default for eligible users. To switch them off, open Settings in the Codex desktop app, go to General, then Composer, and turn off Show predictions. The same toggle turns them back on.

What context do composer predictions use?

Only the conversation in your current thread. If context from past work or connected apps is already in that thread, it can shape a prediction. But predictions don't look up memories or pull from connected apps by themselves.

They work with Codex memory turned off. If memory is on and memory context is already in the thread, it can personalize the suggestion.

One practical detail: predictions can take several seconds to appear, and longer in long conversations.

Do composer predictions cost extra or use my limits?

Not during the beta. OpenAI says predictions are included at no extra cost for eligible Pro users, and generating them doesn't count toward Codex usage limits or consume credits.

The moment you send a message, normal rules apply. That includes an accepted prediction. Sent messages also follow Codex's usual retention and deletion policies.

Are Codex next message suggestions the same as autocomplete?

No. Autocomplete finishes the sentence you're typing. Composer predictions don't do that. During the beta they suggest a complete next message and don't update word by word as you type.

They also differ from the suggestions on Codex's new conversation page. Those help you start a task. Composer predictions only appear inside an existing conversation, after Codex responds.

And on macOS, the feature doesn't need extra keyboard permissions. OpenAI says that if you see a permission request, check which app feature is asking.

Codex is moving on other fronts too. On Windows, Microsoft lists it among the agents that already run inside its MXC agent sandbox. If you're weighing it against Anthropic's approach, see how Claude Code Projects runs parallel threads.

Key facts

  • Announced by OpenAI Developers on October 9, 2026, as a beta.
  • Who: adults 18+ on personal ChatGPT Pro plans, in all supported regions.
  • Where: Codex desktop app, local and SSH threads only. Not in web or desktop Chat.
  • Models: GPT-6 Astra and GPT-6.1 Sol.
  • Press Tab to accept. Accepting never sends the message.
  • Free in beta and doesn't count toward usage limits until you send.
  • Turn off in Settings, then General, then Composer, then Show predictions.

FAQ

How do I accept a composer prediction in Codex?

Press Tab. The suggestion is added to the message box, where you can review or edit it before you send it. Accepting doesn't send it automatically.

How do I turn off Codex predictions?

In the Codex desktop app, open Settings, go to General, then Composer, and turn off Show predictions. Turn it back on in the same place.

Are Codex composer predictions free?

During the beta, yes for eligible Pro users. Generating predictions doesn't count toward Codex usage limits or consume credits. Normal limits and billing apply once you send a message.

Which models support Codex composer predictions?

GPT-6 Astra and GPT-6.1 Sol, in local and SSH threads in the Codex desktop app.

Do Codex composer predictions autocomplete as I type?

No. During the beta they suggest a complete next message after Codex responds. They don't update word by word while you type.

Sources

Keep reading