OpenAI Unveils GPT-5-Codex, Enabling AI to Autonomously Code for Over 7 Hours

OpenAI has announced the release of GPT-5-Codex, a specialized version of its GPT-5 model optimized for "agentic coding" within the Codex platform. This new AI is designed to handle complex software engineering tasks autonomously, with the company reporting instances where the model worked independently for up to seven hours on a single problem. The announcement, made via a tweet from OpenAI, stated, "We’re releasing GPT-5-Codex — a version of GPT-5 further optimized for agentic coding in Codex."

Agentic coding signifies a significant leap in AI's ability to act as an autonomous software engineering partner, moving beyond mere code generation to actively reason, plan, and iterate on tasks. OpenAI's Codex product lead, Alexander Embiricos, highlighted that GPT-5-Codex possesses dynamic "thinking abilities," allowing it to adjust its processing time mid-task. This adaptive approach enables the model to dedicate anywhere from seconds to several hours to a coding challenge, re-evaluating its needs as it progresses.

GPT-5-Codex was specifically trained on real-world software engineering workflows, including building projects from scratch, debugging, refactoring large codebases, and conducting code reviews. OpenAI states that the model outperforms the general GPT-5 on agentic coding benchmarks like SWE-bench Verified and excels in code refactoring. Experienced software engineers evaluating its code review comments found GPT-5-Codex delivered fewer incorrect suggestions and more "high-impact comments."

The new model is now rolling out across various Codex products, making it accessible to a wide range of developers. It is available in the Codex Command Line Interface (CLI), as an IDE Extension for platforms like VS Code, on the web, mobile applications, and integrated for code reviews directly within GitHub. OpenAI confirmed its inclusion for all ChatGPT Plus, Pro, Business, Edu, and Enterprise users, with plans to extend availability to API customers in the future.

This strategic update positions OpenAI more competitively within the rapidly expanding market for AI coding tools, which includes rivals such as Claude Code and GitHub Copilot. While emphasizing the model's power, OpenAI also stresses the importance of human oversight, noting that Codex runs in a sandboxed environment with network access disabled by default to mitigate risks. The company encourages developers to review and validate all agent-generated code before deployment.