Everyone is focused on the benchmark scores, but one number in OpenAI’s GPT-5.6 release caught my attention much more. OpenAI says its internal agentic token usage increased roughly 22x over the last six months, while the share of research compute going toward internal coding inference grew 100x. That feels more significant than another model gaining a few points on a benchmark. The company building frontier models is increasingly using those same models inside the process of building the next...