Gemini 4 Carbon tops graduate-level physics benchmark but loses on agentic tasks

ArtificialAnlys · x · 2026-11-20

Further Artificial Analysis benchmarks of the Gemini 4 lineup: Carbon scores best of the three on CritPt, a graduate-level physics evaluation, but falls behind Neon and Hydrogen on agentic research and knowledge work.

The gap is driven by Carbon's limited improvement at higher reasoning effort levels compared to its siblings.

Original post →

More from Models

Models channel →