As Google and Microsoft successively enhance their AI code writing tools, OpenAI announced the official launch of its software engineering agent Codex (research preview version) in the early morning, bringing a new workflow for developers.
Codex Debut: Cloud Execution, Automated Testing, Multi-task Support
OpenAI stated that Codex is built around the codex-1 model, which is a software engineering-specific version of the o3 reasoning model. It can produce cleaner code, execute instructions more accurately, and has the ability to "self-test until passing". The agent tool runs in a cloud sandbox virtual computer, can connect to GitHub, pre-load the user's code repository, and perform tasks such as adding simple functions, fixing bugs, and testing, with operation times ranging from 1 to 30 minutes.
Moreover, Codex can handle multiple development tasks simultaneously, and users can still operate their computer and browser normally during use, showing its potential as a "virtual development partner".
Limited to Paid Users for Trial, Future Potential Fees
Codex is now officially available for ChatGPT Pro, Enterprise, and Team users to try. OpenAI stated that they will currently provide "generous trial quotas", but will introduce rate limits in the coming weeks, with users able to purchase additional quotas for a fee. Codex will also be expanded to Plus and Edu versions in the future.
Simultaneously, OpenAI has also updated the Codex CLI tool, introducing the optimized o4-mini model and opening it via API, with pricing at $1.50 per million input tokens and $6 per million output tokens.
OpenAI Rumored to Spend $3 Billion to Acquire Windsurf
With the rapid rise of AI coding tools like Cursor, Claude Code, and Gemini Code Assist, OpenAI is also actively expanding its market territory. It is rumored to have completed the acquisition of AI coding platform Windsurf, with an estimated value of $3 billion, forming a two-pronged offensive along with the launch of Codex.
Currently, giants like Google and Microsoft have revealed that about 30% of code is written by AI. The emergence of Codex not only adds value to ChatGPT but may also change the composition of development teams. Josh Tobin, Research Head of OpenAI Agents, stated:
The ultimate goal is to make Codex a true virtual teammate, completing tasks that would take engineers hours or even days.
In the future, as Codex moves towards higher autonomy and deeper integration, the real challenge may not be coding efficiency, but how human engineers will work with this "virtual colleague".
