I'm finding some surprising performance on benchmarks from open weight models when compared to top models https://broskees.github.io/llm-compare/?m=claude-opus-4-6%2C...