OpenAI launches Codex, its new advanced AI coding agent for Pro, Enterprise & Team users
May 16, 2025 at 5:30 PM

OpenAI launches Codex, its new advanced AI coding agent for Pro, Enterprise & Team users

OpenAI has released Codex, a new advanced AI coding agent available in research preview to ChatGPT Pro, Enterprise, and Team users. This tool, accessed from the ChatGPT web app sidebar, acts as an autonomous virtual agent performing tasks such as writing simple features, fixing bugs, running tests, and answering code-related questions through a natural language interface in 1 to 30 minutes. Codex is powered by the specialized 'codex-1' model, a derivative of OpenAI’s o3 architecture, which is optimized for software engineering tasks and produces cleaner code while following precise instructions.

Codex operates on a sandboxed cloud-based virtual machine and integrates with GitHub, enabling repository preloading. It supports multitasking and does not disrupt access to the user's device or browser. The company says Codex can't access the internet for security reasons, helping prevent data leaks and misuse. OpenAI has provided wider initial access, but usage will later be limited by rate caps, with the option to buy extra credits. The company has also updated CodexCLI, its recently launched terminal-based open-source tool, now using the o4-mini model optimized for coding.

While initial autonomy is limited, OpenAI has disclosed a roadmap aiming for Codex to become a fully autonomous software engineer capable of handling independent development workflows. The launch of Codex comes in response to recent advancements from competitors such as Anthropic’s Claude Code and Google’s AlphaEvolve, and it is positioned as a complement to more specialized tools like Cursor and Windsurf Editor, which OpenAI has reportedly recently acquired for around $3 billion.

May 16, 2025 by Mauricio B. Holguin

Ola
Ola found this interesting
ChatGPT iconChatGPT
  378
  • ...

ChatGPT is a generative AI chatbot developed by OpenAI, utilizing the GPT-4o large language model. Launched in 2022, it offers AI-powered conversational capabilities through a web-based platform. Rated 4.4, ChatGPT's core features include its AI-driven responses and chat bot functionality. Notable alternatives include HuggingChat, Google Gemini, and GPT4ALL.

Comments

UserPower
May 16, 2025
2

For very basic documentation and trivial fixes, theses AI assistants are doing just okay. For any medium task, it sucks pretty bad. Way too much noise, it's more annoying than the worse colleagues have met in my entire life. "Vibe coding" is not a thing even if you've got pretty low expectations and it's a waste of time. So I'm waiting for theses large corporations that are selling this idea to really use AI to code a serious project, some project with users and investors. So far, not a single major open-source project has made some progress thanks to theses, and most open-source devs have just wasted time trying to make sense of the generated garbage. And I'm not even against it, I'd be relieve if one day, a program can take care of most of the boring tasks, but I'll believe it when it will work.

Gu