The world: "Can we have Kimi K3 please"? Sberbank: "We have Kimi K3 at home comrade" Sber's latest GigaChat uses MLA + GDN attention 432B, 28B active. They compare the base model (available, MIT) to V4-Flash. meh but stronger than I expected from Russia at this point tbh.
Positive users praise GigaChat 3.5 as Russia's capable SOTA model with reasonable performance, while negative users accuse it of being a stolen Chinese model and criticize Russia.
No Digg Deeper questions have been answered for this story yet.
Well at least I can no longer say that we don't even have Sarvam-tier AI effort.
The world: "Can we have Kimi K3 please"? Sberbank: "We have Kimi K3 at home comrade" Sber's latest GigaChat uses MLA + GDN attention 432B, 28B active. They compare the base model (available, MIT) to V4-Flash. meh but stronger than I expected from Russia at this point tbh.
@teortaxesTex at least it's a SOTA model in russia, and also not that kind of tiny model (below 100B), and deliver a reasonable performance, so not bad for 🇷🇺 huh?
@teortaxesTex It's literally a stolen Chinese deepsi model, they just made a different name for it. That's where the essence of the fascist cancer called "Russia" lies, it steals everything from others and attributes it, calls it its own.