OpenAI introduces GPT-5.1-Codex-Max for long-horizon agentic coding

OpenAI has released GPT-5.1-Codex-Max, a new frontier agentic coding model available in Codex today that replaces GPT-5.1-Codex as the default. Built on an updated foundational reasoning model, it is designed for long-running tasks and operates across multiple context windows through a process called compaction.

• The model natively supports Windows environments and includes training for better collaboration in the Codex CLI.
• It achieves 30% fewer thinking tokens than GPT-5.1-Codex on SWE-bench Verified while maintaining performance.
• A new Extra High (xhigh) reasoning effort is introduced for non-latency-sensitive tasks requiring deeper analysis.
• Internal evaluations show engineers using Codex ship roughly 70% more pull requests, with 95% of OpenAI engineers using it weekly.

The update aims to provide token efficiency and the ability to sustain coherent work over long horizons, enabling project-scale refactors and multi-hour agent loops.