@teortaxesTex About DeepSeek V4 being able to compete with the frontier, I made a new benchmark that suggests open models (particularly the new ultra-sparse ones) are qualitatively worse at problem solving and logical reasoning:

→ View original post on X — @alexjc