it's economy of scales, qwen 3.8 27B regurgitates a lot and that's an intentional tradeoff of space/speed so you teach it patters that can follow for more complex stuff and give it more basic tools that can utilize to build those... the tradeoff comes from the model writting a lot of verbose prose to achieve that.

if the approach works, you have now a viable path to use this complex/hybrid structures where you can pair those chips with gpus and offload whatever is optimum to the model on chip.

Just like you can go and buy a threadripper now (the ram maybe not so much), how much was worth that amount of computing 10 years ago?