Overview

• GPT-5-Codex adjusts its reasoning time per task and has run autonomously for up to seven hours in testing.
• OpenAI reports stronger results on refactoring and SWE-bench Verified, plus substantially fewer incorrect code-review comments compared with GPT-5.
• The rollout integrates a rebuilt Codex CLI, IDE extensions such as VS Code and Cursor, GitHub pull-request reviews, and a cloud agent, and is available to ChatGPT Plus, Pro, Business, Edu, and Enterprise users with API access coming soon.
• New workflow and safety features include image attachments, to-do tracking, clearer tool-call visibility, faster follow-ups via container caching, and default sandboxing with configurable permission modes.
• Executives outlined a vision of large, human-supervised populations of cloud-run agents working continuously in companies’ data centers.