GPT-5.1 Codex Max introduced a fundamental shift in how AI models handle large-scale software development by pioneering context compaction—the first model natively trained to seamlessly operate across multiple context windows, enabling coherent work over millions of tokens in a single task. Rather than relying on traditional chunked processing, this architecture allows the model to plan, implement, and test entire features across sprawling codebases with minimal human intervention, pushing beyond code completion and chat-based suggestions into true autonomous development.

The model demonstrated notable practical advances during extended autonomous sessions, with OpenAI observing continuous operation exceeding 24 hours as it iteratively developed code and resolved test failures without human intervention. Its xhigh reasoning tier achieved 77.9% on SWE-bench Verified while consuming 30% fewer thinking tokens than comparable approaches, and it scored 58.1% on Terminal Bench 2.0, outperforming Gemini and Sonnet at the evaluation period. Available through the OpenAI API and integrated into Microsoft Foundry, the model targets enterprise development teams seeking to automate complex, multi-step coding workflows at scale.