OpenAI has updated its Codex tool with a new version of its language model, GPT-5-Codex. The company says the model can dynamically allocate “think time” and can work on a task from a few seconds to seven hours, allowing it to achieve higher scores in autonomous programming tests.
GPT-5-Codex is already available in Codex products via terminal, IDE, GitHub, and ChatGPT for ChatGPT Plus, Pro, Business, Edu, and Enterprise users. OpenAI plans to open the model to API clients in the future.
The update should make Codex more competitive in a market already dominated by Claude Code, Cursor, and GitHub Copilot. According to the company, GPT-5-Codex outperforms GPT-5 in SWE-bench Verified benchmarks and code refactoring tests on large repositories.
The model has also been trained to conduct code reviews. According to experienced engineers, GPT-5-Codex leaves fewer incorrect comments and adds more important observations.
Codex product lead Alexander Embiricos said the key advantage of the new model is that it does away with the previous "router" system. Instead, GPT-5-Codex decides in real time how long a task will take, and can continue working beyond initial estimates.