You may be correct about the level of models you can actually run on consumer hardware, but it's not fud and you're being needlessly aggressive here.

Incorrect as of a couple of days ago, when Qwen 3.5 came out. It's a GPT 5-class model that you can run at full strength on a small DGX Spark or Mac cluster, and it still works pretty well after quantization.