OpenAI launches GPT-5.2 Codex: How OpenAI is building a smarter AI coding agent
&imwidth=600&imheight=450&format=webp&quality=medium "OpenAI launches GPT-5.2 Codex: How OpenAI is building a smarter AI coding agent")
OpenAI launches GPT-5.2 Codex Photograph: (OpenAI)
OpenAI has launched GPT-5.2 Codex, a specialised AI model built for coding tasks. Designed as a reliable AI coding agent, it handles long, complex projects, reduces errors, and performs better on real-world software benchmarks.
OpenAI has introduced GPT-5.2 Codex, a new version of its AI model designed specifically for coding tasks. The model focuses on improving how AI systems write, understand, and manage code over long and complex projects. The update was announced on OpenAI’s official website, where the company said GPT-5.2 Codex is built to work more reliably as an AI coding agent rather than just a text generator.
What is GPT-5.2 Codex?
GPT-5.2 Codex is a specialised AI model trained to handle software development tasks such as:
Writing new code
Fixing bugs
Reviewing pull requests
Managing large codebases
Trending Stories
&im=FitAndFill=(700,400))
)
)
)
)
&im=FitAndFill=(700,400))
)
)
)
)
)
)
)
)
)
)
)
&im=FitAndFill=(700,400))
)
)
)
)
)
&im=FitAndFill=(700,400))
)
)
)
)
)
&im=FitAndFill=(700,400))
)
)
)
Unlike general-purpose AI models, Codex is designed to follow instructions step by step and stay focused on coding goals for longer periods.
OpenAI said this makes GPT-5.2 Codex more suitable for professional software workflows.
What’s new
OpenAI says GPT-5.2 Codex brings several technical improvements over earlier versions.
According to the company’s release note:
The model performs better on complex coding benchmarks
It shows improved reasoning across long tasks
It reduces common coding errors seen in earlier AI models
OpenAI also stated that GPT-5.2 Codex scored 55.6 per cent on SWE-Bench Pro, a widely used benchmark that measures how well AI models solve real software engineering problems. This score is higher than previous Codex versions.
Built for agent-based coding
One key focus of GPT-5.2 Codex is agent-based coding.
OpenAI explained that the model can:
Work on tasks for several hours without losing context
Handle multiple steps in a single coding job
Operate in sandboxed environments connected to real code repositories
This allows developers to assign tasks such as “fix this issue” or “add this feature” and review the results later.
According to OpenAI, GPT-5.2 Codex is available through:
ChatGPT
Command-line tools
Integrated development environments (IDEs)
Why OpenAI sees coding as a key use case
OpenAI said coding is one area where AI tools are already showing strong real-world value.
The company noted that software development:
Has clear rules
Allows easy testing and verification
Benefits from automation
Because of this, AI models like Codex can assist developers without fully replacing human oversight. OpenAI added that humans remain responsible for reviewing and approving code.
What happens next
GPT-5.2 Codex is part of OpenAI’s broader plan to improve AI-assisted software development.
According to the company, future updates will focus on:
Faster task completion
Better understanding of large projects
Safer deployment in real-world systems
OpenAI said it will continue testing the model with developers and expand access gradually.
