It's because the benchmark is not coding-only.

Gemini models tend to have most knowledge for most domains, and are one of the most intelligent overall. You can check other benchmarks too, on specific categories, those models still beat other SOTA models.

Regarding Opus, Anthropic models often fail to follow instructions, formatting requirements or simply refuse to answer questions (i.e. Fable).

The issue with Gemini models is that they are not as good as using tools or go into weird failure modes when coding or trying to extract/generate specific data. They work amazing, until they don't...