> The last few days, their models feel dumb, thick.

My local Qwen 3.8 27B seems faster and more intelligent than new Luna.