alt Hacker News> you need enormous VRAM laden farms of GPUs to do inference on a model like Opus 4.6.
It's probably a trade secret, but what's the actual per-user resource requirement to run the model?
> you need enormous VRAM laden farms of GPUs to do inference on a model like Opus 4.6.
It's probably a trade secret, but what's the actual per-user resource requirement to run the model?