OpenAI Launches Composer Predictions in Codex

OpenAI has launched a beta feature known as composer predictions for Codex, making it available to ChatGPT Pro subscribers at no added cost. Integrated into the $200-per-month tier, the update allows the system to auto-suggest a user's next instruction based on personal phrasing and session context. The release is currently confined to the ChatGPT desktop app, with no official timeline announced for Codex CLI or editor extension integrations.

The capability is built to accelerate predictable, multi-stage agentic cycles that alternate between code implementation, execution of test suites, error debugging, diff reviews, and pull request commits. Once Codex completes a turn, it generates predicted text in the composer field to let engineers approve or modify next steps without manually writing standard commands. This workflow reflects features found in rival tools such as Claude Code and responds to community feedback outlined in GitHub issue #42587. OpenAI reported that internal testers rated the functionality as one of their favorite additions.

Personalization adapts directly to how an individual writes instructions. An engineer who typically submits concise requests like "run tests" will see succinct suggestions, whereas someone who routinely provides detailed parameters, constraints, and target outputs will receive expanded prompts. Consequently, the feature delivers the greatest efficiency during extended refactoring tasks or bug triage sessions where established task patterns exist, though it offers less utility for brief, exploratory queries.

While the addition promises to cut down on redundant typing, several operational details remain unconfirmed. OpenAI has not published documentation explaining keyboard shortcuts for accepting or dismissing prompts, nor has it clarified how developers can opt out or manage the data used to personalize predictions.