OpenAI launches GPT‑5.3-Codex with faster and broader agentic coding capabilities

OpenAI has announced GPT‑5.3-Codex, the latest update to its family of coding models and the most capable agentic coding agent released to date. This new version combines improvements in coding performance, reasoning ability, and professional knowledge with a significant speed boost: GPT‑5.3-Codex operates 25 percent faster than its predecessor.

Building on these upgrades, GPT‑5.3-Codex now lets users handle continuous, interactive, and long-duration research and execution tasks without losing context. While previous Codex releases focused on code writing and review, this version supports nearly every activity developers and professionals typically perform on a computer, such as debugging, deployment, monitoring, writing product requirements documents, editing copy, user research, and more.

GPT‑5.3-Codex demonstrates its expanded capabilities by achieving new industry highs on the SWE-Bench Pro and Terminal-Bench coding and agentic benchmarks. It also posts strong results on OSWorld and GDPval, establishing its standing in real-world use cases. The model can build functional complex games and applications from scratch over days, further highlighting its multi-phase project competency. GPT‑5.3-Codex is available with paid ChatGPT plans across various platforms, including the app, command line interface, IDE extensions, and web, with API access coming soon.