I would guess such a small difference is within the error margin so "roughly equal" I think is fair to say but different models are good at difference things in

I would guess such a small difference is within the error margin so "roughly equal" I think is fair to say but different models are good at difference things in practice. It is about the same size as GLM 5.3 at about half the price (cached price even less). I believe it's on the intelligence/price pareto frontier. I have tested their Hy3 a bit and it feels okay. That one went from preview to release in less than three months with a substantial increase in benchmarks. But who knows how this space looks like in three months from now.