Even the biggest vision models have traditionally been much smaller than even medium-sized "large" language models, so I don't think the comparison is purely ap

Even the biggest vision models have traditionally been much smaller than even medium-sized "large" language models, so I don't think the comparison is purely apple to apple.