But supposedly they’re the cheapest for certain workloads, especially ones that have high tokens and can make use of caching.

So they’re cutting edge in that way.