A strong, economical Mixture‑of‑Experts (MoE) language model with 236B total parameters (21B activated). Designed for efficient inference, long-context reasoning, and chain-of-thought tasks.
