OpenAI's Internal AI Usage Jumps 22x in Six Months, Compressing R&D Loop

Smart_AI_Hustle · reddit · 2026-08-10

OpenAI's internal agentic token usage has surged roughly 22x over the last six months, while the share of research compute dedicated to internal coding inference grew 100x.

This indicates that frontier models are increasingly being used to build the next generation of models. While not "recursive self-improvement" in a sci-fi sense, the R&D loop is starting to compress: better models help researchers work faster, accelerating the next cycle. The critical question is shifting from how much better the next model is, to how much faster an AI lab can evolve once AI is deeply embedded in its own research process.

Original post →

More from AGI Musings

AGI Musings channel →