I mean, either they cheated on evals ala Llama4, or they have a paradigm that's currently best in class in at least a few standard evals. Both alternatives are

I mean, either they cheated on evals ala Llama4, or they have a paradigm that's currently best in class in at least a few standard evals. Both alternatives are possible, I suppose.