GPT-5.3-Codex expands the coding agent’s remit

OpenAI launched GPT-5.3-Codex on 5 February, combining coding and reasoning improvements in a model intended for longer tasks involving research, tools and execution. The release announcement made it available through paid ChatGPT plans in the Codex app, command-line tool, editor extension and web experience.

The company said API access would follow once it could be enabled safely. That distinction prevented the launch from being an immediate migration option for every application already calling an OpenAI model.

OpenAI described the ability to redirect the agent while it was working without abandoning its context. It also positioned the model for tasks around software development, including analysis and document creation. Its reported speed and benchmark improvements are company claims. The consequential shift was the product’s broader remit: the agent was being asked to carry a sequence of computer work, with user direction continuing during execution. For developers, the initial route remained the Codex product rather than an API integration.