GPT-5.1-Codex-Max System Card

GPT-5.1-Codex-Max is OpenAI's new frontier agentic coding model. It is based on a reasoning model specialized for agentic tasks such as software engineering, math, research, healthcare, and computer use.

In particular, through the compaction process, it natively operates across multiple context windows, allowing it to consistently process millions of tokens in a single task. The model was trained for real-world software engineering tasks such as PR creation, code review, frontend coding, and Q&A.

For safety, both model-level mitigations (training against harmful tasks and prompt injection) and product-level mitigations (agent sandboxing, configurable network access) have been introduced.

According to Preparedness Framework evaluation results, the model shows high capability in the cybersecurity domain but does not reach the 'High' tier.

In the biology domain, it is classified as 'High' capability, applying the same safeguards as GPT-5. On the other hand, in the AI self-improvement domain, it does not reach 'High' capability.