GPT-5-Codex: Autonomous AI Assistant for Developers

OpenAI has introduced GPT-5-Codex, which is a new version of its AI agent designed specifically for programmers. The model is optimized for coding tasks and can operate autonomously for up to 7 hours, performing refactoring, code analysis, and reviews, as well as providing support directly within development environments, IDEs, the terminal, GitHub, and even ChatGPT.

New Features and Flexibility

According to the company, GPT-5-Codex not only writes and completes code but also detects vulnerabilities. The model runs in an isolated environment without network access, but developers can manually enable the internet and external tools. In such cases, the agent can browse pages, create and update projects, and perform code checks on its own.

“Dynamic Thinking”

A key innovation is “dynamic thinking”, the model’s ability to decide on its own how much time and resources are required for a task. For instance, in complex cases, it can continue working for up to 7 consecutive hours.

OpenAI notes that GPT-5-Codex achieves better benchmark results compared to the base GPT-5, particularly in large repositories. Engineers involved in testing reported fewer false remarks and more practical feedback.

Availability and Integration

The model is available to ChatGPT Plus, Pro, Business, Edu, and Enterprise users, with future API integration planned. GPT-5-Codex is already integrated into popular developer environments, including VS Code and GitHub.

To prevent misuse, OpenAI has strengthened its safety systems, as the model possesses advanced knowledge in biology and chemistry. The company has introduced monitoring and blocking of harmful scenarios.