日本語 
OpenAI just dropped GPT-5.1-Codex-Max and the move is pretty clear. Microsoft and Google pushed their coding updates so OpenAI answered with a version built for heavier, longer, messier programming work. The core upgrade is compaction. Instead of choking when the context window fills up the model shrinks’ older chunks of code or conversation so it can keep pushing through larger tasks. That means it can stay on one job for hours or even a full day without losing the thread which is a big deal for system wide refactors or debugging sessions where the input balloon gets out of control.
Also Read: Microsoft, Nvidia and Anthropic Form Major AI Partnership
The model is also leaner. It can hit the same accuracy while burning about 30 percent fewer tokens which matters for people on usage capped plans. The speed bump is real too with processing in many tests jumping between 27 and 42 percent. It even tends to spit out shorter cleaner code for the same outcome which usually means better structure and easier maintenance. OpenAI also tightened security on long running agents and trained the model to behave better in Windows which makes sense given the ecosystem shift. API access is coming soon and the model starts rolling out to paid plans right away.
English