@nrehiew_ Hey, just made a logic reasoning / problem solving benchmark where open weights models get completely lost, but the frontier models make it look easy. Curious about your hypothesis why, thinking it's sparsity related:

→ View original post on X — @alexjc,