OpenAI Debuts GPT‑5.1-Codex-Max Coding Model and It Already Completed a 24-Hour Task Internally

Why It Matters

Codex‑Max raises the ceiling for AI‑assisted software engineering by delivering higher accuracy, token efficiency and persistent, multi‑hour coding sessions, potentially accelerating development cycles and giving OpenAI a competitive edge over rivals like Google in the enterprise coding‑assistant market.

Summary

OpenAI has launched GPT‑5.1‑Codex‑Max, a new agentic coding model that replaces GPT‑5.1‑Codex as the default in its Codex developer environment. The model introduces a compaction mechanism that enables long‑horizon reasoning across millions of tokens, cutting token usage by about 30% and allowing it to complete tasks that run for more than 24 hours. Benchmark results show Codex‑Max leading or matching Google’s Gemini 3 Pro on key coding tests, with 77.9% accuracy on SWE‑Bench Verified and 58.1% on Terminal‑Bench 2.0. The model is currently available through the Codex CLI and to ChatGPT Plus, Pro, Business, Edu and Enterprise users, with a public API slated for future release.