Post

can a LLM knower explain to me what changes between "model names" (eg, Sonnet -> Opus -> Fable) and what changes between "model versions" (eg, 5 -> 5.5)? Like do the names represent fundamentally different parameter counts and the versions are more incremental improvements?

1 repost5 likes

yeah roughly, although i would describe it more as size classes with intended usecases and pricing, these are fundamentally marketing terms that terra sort of disappeared from the luna-terra-sol-astra lineup had mostly to do with there not being a clear enough market for terra

what Laurens said model versions also usually come with some architecture change, like Sol 6 being way cheaper than 5.6 indicates it doesn't work the same way. That said, it's usually just a rule of thumb (e.g. Sol 6.1 is probs a quantization of Astra) Name -> category/class, version -> generation