Last updated: October 10, 2026

Codex composer predictions: the next message, suggested

On October 9, 2026 OpenAI put composer predictions into beta in Codex. After Codex finishes a response, it can show a suggested next message in faded text. Press Tab to put it in the box, review it, and send it yourself.

Short version: it is a Pro-only beta in the Codex desktop app (local and SSH threads), on GPT-6 Astra or GPT-6.1 Sol. Predictions are on by default and do not use Codex credits, but the messages you send still count toward normal usage limits.

Who can use it

Adults aged 18 and over on a personal ChatGPT Pro plan, in supported regions.
The Codex desktop app only, in local and SSH threads.
Thread model must be GPT-6 Astra or GPT-6.1 Sol.
Not available in Chat on the web or desktop. Other plans are excluded during the beta.

OpenAI’s announcement said “for Pro users”. Its Help Center article gives the eligibility above, so check your plan before you look for the feature.

Source: OpenAI Help Center: Composer predictions in Codex

Source: OpenAI Developers announcement (Oct 9, 2026)

How to accept a suggestion

Finish a Codex turn. A suggestion may appear in the message box as faded text; not every response gets one.
Press Tab to add it to the message box. This does not send it.
Edit it if needed, then send it yourself.
Ignore it and type your own message at any time.

Source: OpenAI Help Center: Composer predictions in Codex

How to turn it off

Predictions are on by default for eligible users. To turn them off, open Settings, go to General, then Composer, and switch off “Show predictions”. You can turn it back on in the same place.

Source: OpenAI Help Center: Composer predictions in Codex

Cost during the beta

Generating predictions does not count toward Codex usage limits or credits.
Sending a message, including an accepted prediction, follows normal usage limits and billing.

Source: OpenAI Help Center: Composer predictions in Codex

What it can see

Predictions use the conversation in the current thread.
They do not look up memories or connected apps on their own.
If memory is on and already in the thread, it can shape predictions.
Messages you send follow Codex’s retention and deletion policies.

Source: OpenAI Help Center: Composer predictions in Codex

Limits OpenAI lists

Predictions do not appear after every response, and some are skipped.
They can take several seconds to appear, longer in long conversations.
They suggest one complete message and do not update as you type.
They never send messages or continue tasks on their own.
They may not match what you meant, so read them before you send.

Source: OpenAI Help Center: Composer predictions in Codex

FAQ

What are Codex composer predictions?

A beta feature in the Codex desktop app that suggests your next message after Codex finishes a response. Press Tab to add it to the box, then review and send it yourself.

Can I use it on the Plus plan?

OpenAI’s Help Center limits the beta to personal ChatGPT Pro plans, so other plans are excluded for now.

Does Tab send the message?

No. Tab only adds the suggestion to the message box. You send it yourself.

How do I turn predictions off?

In Codex, go to Settings, then General, then Composer, and turn off “Show predictions”.

Do predictions use my Codex credits?

Generating predictions does not count toward Codex usage limits or credits. Messages you send, including accepted predictions, follow normal usage limits and billing.

Independent guide, not affiliated with OpenAI. Facts come from OpenAI’s Help Center article and developer announcement, checked on October 10, 2026. The feature is in beta and may change.