GPT-5-Codex
GPT-5-Codex, a GPT-5 version tuned for agentic coding and code review, becomes the default engine across Codex CLI, IDE, cloud and mobile.
Spends variable thinking time (quick on simple tasks, hours on large ones), is trained for code review, and raised OpenAI's internal refactoring eval from 33.9% (GPT-5 high) to 51.3%. API release followed on 2025-09-23. Same day, Codex was unified into one product tied to a ChatGPT account.
- Date
- Monday, 15 September 2025
- Lab
- OpenAI
- Kind
- model
- Access
- closed API
- Price
- API from 2025-09-23
Figures
| Measure | Value | Measured by |
|---|---|---|
| Internal code-refactoring eval | 51.3% GPT-5 (high) 33.9%; proprietary eval, details undisclosed | company |
OpenAI reported Codex working independently for more than seven hours on large tasks (Simon Willison summary).
Sources
- openai.com/index/introducing-upgrades-to-codex/
- simonwillison.net/2025/Sep/15/gpt-5-codex/
- developers.openai.com/api/docs/changelog
- deploymentsafety.openai.com/gpt-5/
This record was checked against its sources on 6 October 2026. How we check