I got am m2 max with 96gbyte ram. I bought it to create photogrammetry, which i havent really used it with it because my focus changed. But i do regularely run my own llms (120B-q4) which work at acceptable speeds (7-8 tokens/s for llama3.2 90b q4km, 14 t/s for quen 2.5 q4km). Like that ram usage goes up to 83gbyte.