OpenAI has announced the release of GPT-5-Codex, a new model designed to take AI coding assistance beyond autocomplete and into the realm of autonomous software engineering.
The upgraded system builds on GPT-5 but is specifically optimized for what OpenAI calls “agentic coding”—where the AI acts more like a teammate, capable of managing complex tasks with minimal human input.
Key features
Real Engineering Capabilities: GPT-5-Codex can tackle end-to-end projects, refactor large codebases, fix bugs, write tests, and conduct code reviews.
Dynamic Effort: The model adjusts how much “thinking time” it spends depending on the difficulty of the task—finishing small fixes quickly while persisting through challenges for hours on larger assignments. In testing, it ran autonomously for more than seven hours on complex jobs.
Smarter Code Reviews: The system doesn’t just scan code superficially. It runs tests, reasons through dependencies, and provides detailed review comments that developers have rated as more accurate and actionable than previous versions.
Availability
GPT-5-Codex is available immediately across OpenAI’s Codex ecosystem, including terminal access, IDE extensions, cloud and web environments, GitHub integrations, and the ChatGPT mobile app. It is now the default model for code reviews and cloud-based coding tasks.
Why It Matters
Unlike earlier AI coding tools that required frequent prompts and close supervision, GPT-5-Codex allows developers to give high-level instructions while the model figures out execution details. This reduces repetitive manual work, boosts efficiency, and makes the AI feel less like a tool and more like a collaborative partner.
The release signals OpenAI’s push to redefine how developers interact with AI in the coding workflow—turning it into an independent contributor rather than just an assistant.
