OpenAI introduces Codex — a cloud-based AI agent for programming

OpenAI has announced Codex, a new AI agent designed to assist with programming tasks.

Built on the codex-1 model, which leverages the “reasoning” version of GPT-4 (o3), Codex is optimized to write clean, tested code. It can independently test and debug programs until they reach a fully functional state.

Codex runs in the cloud, minimizing browser load, and works even on smartphones. It can perform multiple tasks simultaneously, connect to GitHub repositories, and automatically generate a log of all code changes upon completion. The model supports a context size of up to 192,000 tokens.

Currently, Codex is available to Pro, Enterprise, and Team subscribers, with Plus and Edu users gaining access later. The AI will be integrated into the ChatGPT sidebar and accessible via a separate window. In the coming weeks, OpenAI plans to introduce limits on agent speed and a system for purchasing additional credits.

OpenAI has also introduced a lighter version — codex-mini, powered by the o4-mini model, available through the API at a cost of $1.5 and $6 per million input and output tokens, respectively.