A suggested follow-up after a response
OpenAI’s 9 October release note introduces Composer predictions in Codex. This is separate from the earlier follow-up steering change: it proposes text for your next message after Codex responds.
Who can use the beta
The beta documentation limits access to adults on personal ChatGPT Pro accounts in supported regions, using the updated Codex desktop app. Local and SSH threads support predictions with GPT-6 Astra or GPT-6.1 Sol.
Pressing Tab accepts suggested text into the composer. You can change it before sending; acceptance never sends it automatically. Predictions can be skipped, so their absence after one response does not establish a fault.
Controls, context and usage
Predictions start enabled for eligible accounts. The switch is in Settings → General → Composer → Show predictions. The feature uses the current thread; it does not independently retrieve memories or connected-app information.
During the beta, prediction generation adds no charge and consumes neither Codex credits nor usage allowance. Sending the resulting message still follows ordinary Codex limits and billing. This describes beta generation, not unlimited agent work.
Keep the buying decision separate
This announcement supplies no comparison of code quality or review effort. The coding assistant task worksheet records those questions against your actual backlog. The Codex profile and Cursor/Codex comparison provide related context.
Sources and scope
Source review only. Account eligibility, model access and live behavior were not independently exercised. No recommendation, score or winner changed.