OpenAI just rolled out GPT-5 Codex, and it’s now available in Cursor.
Codex is a specialized variant of GPT-5 optimized for coding, refactoring, and code review. It’s designed to handle long, complex tasks autonomously (sometimes working for hours), adapt reasoning depth based on task complexity, and catch critical issues in reviews.
Benchmarks: 74.5% on SWE-bench Verified
OpenAI also highlithed that it uses drastically fewer tokens for simple tasks so hopefully it is also cheaper to use
No information on context length, probably same as GPT-5 so roughly 200k