LLM direct response models are quite mature, yes (4o)

LLM based MoE architectures with some kind of reasoning process ( Claude 3+, o series, R1, grok 3 with thinking ), are the equivalent of v0.2 atm, and they're showing a lot of promise.