OpenAI launches GPT-5 Codex, a coding-optimized AI model
Readers help support Windows Report. We may get a commission if you buy through our links.
Read our disclosure page to find out how can you help Windows Report sustain the editorial team. Read more
After announcing GPT-5 last month, OpenAI has followed up with the launch of GPT-5 Codex. As tge name suggests, this model is built specifically for coding and software engineering. You can expect it to handle more practical developer tasks, ranging from starting new projects to managing large-scale refactoring.
OpenAI highlights that on SWE-bench Verified, GPT-5 High scored 72.8%, while Codex surpassed with 74.5%. The performancs are even more striking when we consider refactoring. In that test, GPT-5 Codex reached 51.3%, far surpassing GPT-5’s 33.9%.
One of GPT-5 Codex’s standout features is its ability to adjust reasoning time based on task complexity. In OpenAI’s testing, the model worked autonomously for over seven hours on large-scale engineering problems.
Front-end developers will likely also benefit from this model as it improves workflows. Codex can take images as input, visually track its progress, and generate screenshots of completed tasks. That’s quite handy for web and mobile projects.
OpenAI says Codex does a pretty good job when it comes to code reviews. The model can now analyze pull requests against actual diffs, review intent versus implementation, and even run code with tests to validate functionality. Internally, OpenAI reports that GPT-5 Codex is already flagging hundreds of issues every day.
If you are a developers, you can start using GPT-5 Codex today via ChatGPT Plus, Pro, Business, Edu, and Enterprise plans, as well as via the Codex CLI, IDE extensions, web, and mobile. GitHub users also benefit, with Codex powering code reviews natively. Moreover, we expect OpenAI to soon offer API access for those integrating via Codex CLI keys.
