OpenAI upgrades Codex with a new version of GPT-5 (techcrunch.com)

OpenAI announced GPT-5-Codex, a new GPT-5 variant integrated into its Codex coding agent and rolling out across Codex products (terminal, IDE, GitHub, and ChatGPT) for ChatGPT Plus, Pro, Business, Edu and Enterprise users, with API access promised later. The upgrade is aimed squarely at competing in a crowded AI-coding market (Claude Code, Cursor, GitHub Copilot) and, per OpenAI, yields measurable gains on agentic coding benchmarks and long-form refactoring tasks drawn from large repositories. Technically, GPT-5-Codex introduces dynamic, real-time compute allocation for problem-solving: rather than using a static router to assign a model and compute budget at the outset, the model can extend its “thinking” time mid-task—ranging from seconds up to reported cases of seven hours—to pursue deeper exploration. OpenAI says this improves agentic coding performance (SWE-bench Verified) and produces better code-review feedback: engineer evaluations found fewer incorrect comments and more high-impact suggestions. The change implies more adaptable, persistent agent behavior for complex coding workflows, which could raise expectations for automated refactoring, review quality, and long-running program synthesis in developer tools.