OpenAI Rolls Out GPT-5-Codex, a New AI Model to Take On GitHub Copilot and Cursor

OpenAI has launched new PT-5-Codex, a coding-focused version of its GPT-5 model that now powers all of its Codex tools. The release comes at a time when the market for AI coding assistants is heating up, with rivals like GitHub Copilot, Cursor, Claude Code, and Windsurf attracting strong developer interest.

Unlike earlier models, GPT-5-Codex doesn’t just spit out quick answers. It can adjust how long it “thinks” depending on the task, says OpenAI. That means developers can get instant help for small fixes or let the model run independently for seven hours or more on big projects such as refactoring or debugging.

OpenAI says that it has also trained the model to act as a code reviewer. It can analyse entire repositories, run tests, and surface flaws before code is shipped. Engineers who tested it said the system made fewer irrelevant comments and highlighted more serious issues compared with earlier versions.

In a key benchmark for real-world engineering tasks, GPT-5-Codex scored higher than GPT-5. It also showed major improvements in large-scale refactoring across languages including Python, Go, and OCaml suggesting cleaner, more maintainable code with less back-and-forth.

Codex has recently been unified into a single product tied to ChatGPT accounts, making it easier to move between local and cloud setups. Developers will have a variety of convenient options to use AI to help them with coding, regardless of their preferred working environment. They can also attach screenshots, create to-do lists, and request GitHub pull request reviews directly within the tools.

OpenAI says GPT-5-Codex runs in a sandboxed environment with network access off by default, reducing risks such as malicious code execution. GPT-5-Codex is available to ChatGPT Plus, Pro, Business, Edu, and Enterprise users, with API access expected later this year.

Read More: OpenAI and Nvidia to Announce UK Data Center Push; India in Line for 1GW AI Hub