Yup. I'm done with US companies. Let's go China !
replyhave fun with your sub-tier models then. More compute for me
reply...until you get rug-pulled (like with Fable recently) because the model you're depending on is proprietary, then you have fun with no model at all.
replyThe sooner Chinese labs catch up the better it is for all of us, even if you don't use their models, as they're the ones who define the baseline capability that cannot be taken away from you (no one is going to limit/remove access to an LLM if you can get equivalent/better unrestricted access from Chinese open-weight models).
Tried Xiaomi MiMo v2.5 via opencode today. Since Sonnet 5's release week, Sonnet 4.6 has been feeling like a vegetable, with Sonnet 5 itself being only a little better. MiMo on the other hand feels like Sonnet 4.6 did up until very recently. Absolutely impressive.
replyIn some ways, more impressive than GPT 5.5 with high(!) thinking. GPT says quite some nonsense from time to time; didn't see any sign of this in MiMo so far, which is a pretty wild difference.