Codex Composer Predictions Beta Suggests the Next Message in Desktop Threads
OpenAI added composer predictions to Codex in beta. Eligible personal ChatGPT Pro users aged 18 and older can see complete next-message suggestions in the Codex desktop app for supported local and SSH threads, accept them with Tab, edit before sending, and turn the feature off in Settings.
OpenAI has added composer predictions to Codex in beta: a suggestion for the user's next complete message can appear after Codex finishes a response in an eligible desktop thread. The feature is available for adults aged 18 and older on personal ChatGPT Pro plans, in all supported regions, in the Codex desktop app. (OpenAI Help Center)
The important product detail is control. A prediction does not send anything on its own: users press Tab to accept the suggestion into the message box, review or edit the text, and send it themselves when ready. (OpenAI Help Center)
Key Takeaways
- Beta scope: composer predictions are in beta for adults 18+ on personal ChatGPT Pro plans, in all supported regions, in the Codex desktop app. (OpenAI Help Center)
- Thread support: predictions are available in local and SSH threads when a supported model is selected. (OpenAI Help Center)
- Model support: OpenAI lists GPT-6 Astra and GPT-6.1 Sol as supported models. (OpenAI Help Center)
- Not every turn: after Codex finishes responding, a suggestion may appear, but the composer can skip predictions and a prediction does not appear after every response. (OpenAI Help Center)
- Accept is not send: Tab accepts the text into the message box so the user can review or edit it before sending. (OpenAI Help Center)
- Beta usage treatment: during beta, predictions are included at no additional cost for eligible Pro users, and generating predictions does not count toward Codex usage limits or consume credits. (OpenAI Help Center)
- Surface limits: predictions are not available in Chat on web or Desktop, and the Help Center says Work predictions may appear when using local Work threads and conversations. (OpenAI Help Center)
- User setting: users can turn predictions off in the Codex desktop app under Settings, General, Composer, Show predictions. (OpenAI Help Center)
How Composer Predictions Work
OpenAI describes composer predictions as next-message suggestions based on the conversation in the current Codex thread. The Help Center says a suggestion may appear in the message box after Codex finishes responding, but predictions may be skipped and do not appear after every response. (OpenAI Help Center)
The suggestions are complete next messages, not word-by-word completions while the user types. OpenAI also says users do not need to wait for a prediction before typing their own message. (OpenAI Help Center)
The acceptance flow is deliberately manual. Pressing Tab accepts a suggestion into the message box, but the user still reviews or edits the text and sends it when ready; accepting the prediction does not send the message. (OpenAI Help Center)
Availability and Limits
For now, the availability box is narrow. The feature is in beta for adults aged 18 and older on personal ChatGPT Pro plans, in all supported regions, in the Codex desktop app. OpenAI says predictions are on by default for eligible users. (OpenAI Help Center)
Thread and model support are also specific. Predictions are available in local and SSH threads when the user selects GPT-6 Astra or GPT-6.1 Sol. (OpenAI Help Center)
The feature is not a Chat-wide autocomplete system. OpenAI says predictions are not available in Chat on web or Desktop, and its Work wording is limited: "In Work, predictions may appear if you use local Work threads and conversations." (OpenAI Help Center)
Cost, Usage, and Settings
During the beta, OpenAI says predictions are included at no additional cost for eligible Pro users. Generating predictions does not count toward Codex usage limits or consume credits, but normal Codex usage limits and billing apply when the user sends a message, including an accepted prediction. (OpenAI Help Center)
Users can disable the feature from the Codex desktop app by opening Settings, going to General, then Composer, and turning off Show predictions. The same setting turns predictions back on. (OpenAI Help Center)
What to Watch
Treat the beta as a controlled desktop feature, not a general ChatGPT launch. The safest rollout advice is to verify the account, app, thread type, and selected model first, then make sure users understand that Tab only stages a suggestion for review and editing. The user still decides what gets sent. (OpenAI Help Center)
Sources
- OpenAI Help Center, "Composer predictions in Codex": https://help.openai.com/en/articles/20001601-composer-predictions-in-codex
- OpenAI Developers announcement post, October 9, 2026: https://x.com/OpenAIDevs/status/2108624138369929725
- OpenAI Developers follow-up with Help Center link, October 9, 2026: https://x.com/OpenAIDevs/status/2108624150344646752
Read next
Related coverage- OpenAI
GPT-6.1 Sol: Astra-Level Coding at a Fifth of the Price, With Cost at Every Effort Level
OpenAI released GPT-6.1 Sol on September 29, 2026, at $2 input and $10 output per million tokens, a fifth of GPT-6 Astra's price, with cached input at $0.10. On OpenAI's charts it edges past Astra's best DeepSWE score at high effort and trails Astra by 2.1 points on OSWorld offline. The API model ID is gpt-6.1-sol.
- ChatGPT
ChatGPT Pro 500 Launches at $500 With 25x Plus Usage and Astra Ultrafast, While Pro 200 Drops to 10x
OpenAI's Pro 500 costs $500 a month for 25 times ChatGPT Plus usage plus Ultrafast, which now covers GPT-6 Astra and GPT-6.1 Sol and uses included limits at 8x the Standard rate. Astra Ultrafast generates tokens up to 8x faster in Codex. Pro 200 reopens; Sottiaux and The New Stack say it falls from 20x to 10x Plus.
- Codex
Codex Permission Profiles: Least-Privilege Controls for Local Agent Work
OpenAI shipped Codex permission profiles in beta -- reusable, inheritable policies that replace the coarse sandbox_mode/sandbox_workspace_write combo. A profile binds OS-enforced filesystem read/write/deny rules (down to **/*.env) to per-domain network and Unix-socket rules. Enterprise admins get fail-closed allowlists via requirements.toml. Profiles govern local sandboxed command execution only, not MCP servers, app connectors, browser, or cloud.
Frequently Asked Questions
Who can use Codex composer predictions?
OpenAI says composer predictions are in beta for adults 18 and older on personal ChatGPT Pro plans, in all supported regions, in the Codex desktop app. The Help Center says they are not available on other plans on the web or in Chat.
Which Codex threads and models support predictions?
Predictions are available in local and SSH threads when the user selects a supported model. OpenAI lists GPT-6 Astra and GPT-6.1 Sol as the supported models.
Does pressing Tab send the predicted message?
No. Pressing Tab accepts the suggestion into the message box. The user can review or edit the text and then send it when ready; accepting a prediction does not send the message.
Do composer predictions count against Codex usage limits?
During the beta, OpenAI says predictions are included at no additional cost for eligible Pro users, and generating predictions does not count toward Codex usage limits or consume credits. Normal Codex usage limits and billing apply when a message is sent, including an accepted prediction.