OpenAI releases GPT-5-Codex, a GPT-5 version tuned for agentic coding

OpenAI/Codex

OpenAI released GPT-5-Codex, optimized for end-to-end software engineering tasks including refactoring, debugging, and code review. It outperforms earlier GPT-5 variants on benchmarks, scoring 74.5% on SWE-bench Verified (vs. 72.8% on GPT-5 High) and 51.3% on refactoring tasks, and is available via ChatGPT Pro, Business, and Enterprise plans with scalable usage limits.

• Trained on real software-engineering work; runs independent tasks for up to 7 hours with adaptive compute allocation.
• Integrated IDE extensions for VS Code, Cursor and similar tools; rebuilt CLI and GitHub PR integration.
• Available to ChatGPT Pro, Business, Edu and Enterprise users with usage limits scaled by plan.
• Cloud sandbox environment preloaded with code for reproducible, isolated task execution.