Post-training can have big gains no? I don't think the current sizes at ~1T are saturated in intelligence (it's like saying AlphaGo Master is just a post-trained version of AlphaGo Lee)