Comment by armen52

I don't understand this assertion, but maybe I'm missing something?

Google included a SWE-bench score of 63.8% in their announcement for Gemini 2.5 Pro: https://blog.google/technology/google-deepmind/gemini-model-...