OpenAI has released GPT-5.1-Codex-Max, a new model optimized for agentic coding tasks that is now the default in the Codex CLI. Unlike the general-purpose GPT-5.1, this model is natively trained to operate across multiple context windows using a process called compaction.
- It achieves a 76.5% score on SWE-Bench Verified at "high" thinking level and 77.9% at "xhigh," surpassing Gemini 3 Pro and Claude Sonnet 4.5.
- The model scores 58.1% on Terminal Bench 2.0, outperforming both Gemini 3 Pro (54.2%) and Claude Sonnet 4.5 (42.8%).
- Compaction allows the model to prune history while preserving important context, enabling it to complete tasks that previously failed due to context-window limits.
OpenAI also announced that GPT-5.1 Pro is rolling out to all ChatGPT Pro users, with GPT-5 Pro being retired in 90 days.