According to ChainCatcher, citing Jintian's report, OpenAI has released a brand new cloud-based code Agent called Codex on its official website. Codex can automatically perform various tasks for developers, such as writing functional code, answering questions about code libraries, fixing errors, and proposing pull request reviews, with each task running in its own cloud sandbox environment, while also being able to preload the user's code library.
It is reported that Codex is powered by the Codex-1 model, which is a fine-tuned version of OpenAI-o3, specifically optimized for software engineering. Through reinforcement learning in multiple environments, the model is trained on real-world coding tasks, generating code that aligns with programmers' coding styles and pull request preferences, and strictly follows instructions.
