I'm pretty happy about the recent releases too, especially the larger models, even though I "only" have 32GB GPUs.

My attitude is, if AI Winter falls tomorrow, whatever is available now might be all we get to have thereafter, at least until the open source community acquires the hardware to advance the technology ourselves.

The hardware will trickle down into our hands via the second-hand market, eventually. The difference between technology that costs as much as a luxury sedan and technology that costs as much as a burrito is about eight years.

With such powerful models as GLM-5.1, MiMo-V2.5-Pro, and MiniMax-M2.7 available now, even if all advances stop right here, we'll be in a really happy place for many years to come. As better hardware becomes available, these more powerful models will be ours to use on that hardware.

We would also be able to leverage these larger models to make better small models via distillation, so if some of us get more powerful hardware and the rest lag behind with 12GB or 24GB GPUs, as our datasets and distillation pipelines improve, so should the models which will fit in those smaller GPUs.