GigaChat opens a 432B reasoning model

Sber’s GigaChat 3.5 Reasoning release puts a 432B-parameter mixture-of-experts model on Hugging Face, with 28B active parameters, a 262K-token context, and MIT-licensed FP8 weights, according to the published model card. That is a serious open-weight artifact but not a free laptop deployment: its own launch example uses eight H100s, so teams should benchmark the 28B-active routing economics before treating it as a frontier API substitute. For teams tracking open-weight adoption’s narrow denominator, file this as more supply—not proof of production demand.