On September 16, OpenAI launched GPT-5-Codex, a model optimized for software engineering, enhancing Codex’s Agentic Coding abilities. This new version dynamically adjusts thinking time based on task complexity, autonomously handling challenging tasks for up to seven hours. Benchmark tests demonstrate a performance improvement over GPT-5, particularly in code accuracy, achieving 74.5% versus 72.8% in key metrics.
GPT-5-Codex is now the default tool for cloud tasks and code reviews, offering seamless integration through Codex’s CLI and IDE for developers. It significantly reduces incorrect comments during code reviews, with only 4.4% deemed inaccurate, compared to GPT-5’s 13.7%. Codex can now also scan development intentions to detect critical flaws, enhancing defect identification.
As competition intensifies among AI programming tools like Claude and GitHub Copilot, OpenAI’s enhancements improve programming automation and collaborative capabilities, positioning Codex as a powerful asset for developers.
Source link
