OpenAI has rolled out a major upgrade to its AI coding agent, Codex, with the release of GPT-5-Codex, a new model designed to tackle complex coding problems more intelligently and flexibly. Unlike its predecessors, GPT-5-Codex dynamically adjusts how much “thinking” time it spends on a task—ranging anywhere from a few seconds to as long as seven hours—making it more effective on real-world programming challenges.
The new model is already available to ChatGPT Plus, Pro, Business, Edu, and Enterprise users across Codex interfaces, including terminal, IDE integrations, GitHub, and ChatGPT itself. OpenAI says API access will follow soon, opening the door for broader adoption.
Smarter, more agentic coding
OpenAI claims GPT-5-Codex significantly outperforms GPT-5 on SWE-bench Verified, a benchmark that measures “agentic coding” abilities, and on large-scale code refactoring tasks drawn from real repositories. The model has also been evaluated for code review quality, with professional engineers reporting that GPT-5-Codex produces fewer incorrect comments while providing more “high-impact” feedback compared to earlier versions.
In a press briefing, Alexander Embiricos, OpenAI’s Codex product lead, explained the breakthrough:
“GPT-5-Codex doesn’t rely on a router to pre-determine how much compute or time to spend on a problem. It can dynamically decide—five minutes in—that it needs an hour more, or even several hours, to work through the task.”
This flexible approach stands in contrast to models that assign computational effort upfront, giving Codex an edge on problems that unfold unpredictably during execution.
A crowded AI coding landscape
The launch comes at a time when the AI coding market is heating up. Competitors like Claude Code, Anysphere’s Cursor, and Microsoft’s GitHub Copilot are all pushing hard for developer adoption. Cursor alone surpassed $500 million ARR earlier in 2025, while rival editor Windsurf became the subject of a high-profile acquisition battle that split its team between Google and Cognition.
With GPT-5-Codex, OpenAI is signaling that it doesn’t just want to keep pace—it wants Codex to set the bar for long-horizon, agentic coding performance.
What’s next
For now, GPT-5-Codex is reserved for ChatGPT’s higher-tier users, but OpenAI’s roadmap includes API availability that could bring the upgraded coding assistant into custom developer workflows and third-party platforms. Given its ability to spend hours refining solutions or reviewing codebases, it’s likely to appeal to software teams working on large, mission-critical repositories where accuracy and deep reasoning matter most.
