It is on top for many benchmarks, only not the coding/agentic ones.

Still one of the most intelligent models overall, most likely to get any question you ask correctly (without tools).