Comment by Degorath

I sort of agree. Any benchmark that shows gpt-6 sol as better than gpt-5.6 sol is incredibly suspect.

Or I'm wrong and working with a model has an aspect to it that is not captured by any benchmark.