AliceAI Foundation 80B-A3B

AliceAI-Foundation-80B-A3B-Base is Yandex's first large language model trained fully from scratch with no third-party components, open-sourced 2026-09-21 under Apache 2.0. It is a text-only Mixture-of-Experts model with 80B total parameters and ~3B active per token (512 experts, top-10 routing plus 1 shared expert) using a hybrid attention design that interleaves linear (KDA-style) attention blocks with gated full-attention blocks. Native 262,144-token context. Released as a pretrained (base) checkpoint intended as a foundation for further fine-tuning and research rather than direct production chat; no instruct variant at launch.