GPT-5.1 Codex (high) is a specialized version of OpenAI's GPT-5.1 family, specifically optimized for advanced programming and technical reasoning. Released in November 2025, it represents a mid-cycle upgrade over the initial GPT-5 series, focusing on improved instruction following, agentic steerability, and a more natural communication style. The "high" designation refers to the model's reasoning effort setting, which allocates additional compute during inference to solve complex logic and software engineering problems.
Built with an agentic focus, the model is designed to handle long-running tasks such as multi-step code refactors and autonomous debugging. It features a 400,000-token context window, allowing it to ingest large-scale repositories and maintain consistency across complex development workflows. It is also natively multimodal, possessing the ability to process and reason over visual technical documentation, diagrams, and user interfaces alongside traditional source code.
Benchmark results show significant performance in technical evaluations, including high scores on the SWE-bench for verified software engineering tasks. The model utilizes a dynamic reasoning stack that can be adjusted via API parameters, with the "high" setting specifically targeted at high-stakes engineering where precision is prioritized. It also incorporates "compaction" techniques to manage its history and preserve critical information during sessions that approach the context limit.