OpenAI announced GPT-5.1-Codex-Max on 19 November 2025, describing it as a specialized coding model with expanded context windows and improved multi-file reasoning [source]. The provider states the model handles up to 256,000 tokens and supports "repository-level understanding" for complex codebases.

The changelog notes several architectural changes from the GPT-4 series. GPT-5.1-Codex-Max uses a modified attention mechanism optimized for structured text, according to OpenAI's technical documentation. The model was trained on code repositories through October 2025, extending the knowledge cutoff beyond GPT-4's September 2021 date.

OpenAI reports benchmark improvements on HumanEval, MBPP, and internal multi-file editing tasks. The provider claims 89.2% pass@1 on HumanEval, compared to 67% for GPT-4 and 84.1% for the earlier GPT-4.5-Codex variant released in March 2025. No independent verification of these figures has been published.

The model is available through OpenAI's API at a higher tier than standard GPT-4 access. Pricing is set at $0.12 per 1,000 input tokens and $0.48 per 1,000 output tokens, approximately four times the cost of GPT-4 Turbo. OpenAI states the model is intended for "professional development workflows" and may not be suitable for general conversational use.

The announcement includes no mention of safety evaluations specific to code generation, such as vulnerability introduction rates or license compliance checks. OpenAI's changelog does not specify whether the model was trained on proprietary code repositories or only open-source data.