GPT-5.1-Codex-Max is a new OpenAI model aimed at developers.
OpenAI has launched a new code model for developers. GPT-5.1-Codex-Max replaces the existing Codex variant in the full Codex environment and focuses on long-term, complex development processes. The focus is on reliability, deeper reasoning, and real-time interaction.
Stronger Performance than Google Gemini 3 Pro
The timing is noteworthy: Google just yesterday unveiled its powerful Gemini 3 Pro model. Yet, Codex-Max performs slightly better on key benchmarks. On SWE-Bench Verified, the model achieves 77.9 percent accuracy, compared to 76.2 percent for Gemini 3 Pro. OpenAI also leads in Terminal-Bench 2.0 with 58.1 percent and performs similarly to Gemini in other benchmarks.
One of the biggest innovations is the compaction system. This allows Codex-Max to continue working across millions of tokens without context loss. Unimportant details are automatically removed, while relevant information is retained. OpenAI states that internal testing sessions could last more than 24 hours.
This efficiency also translates into lower costs: for average reasoning tasks, Codex-Max uses up to 30 percent fewer tokens than its predecessor.
Deeper Integration into Codex Environments
The model is already available in Codex CLI and in interactive code environments. This makes it possible not only to write software but also to perform simulations and visualizations. Currently, there is no public API, but it will follow later.