Qwen3 Next 80B A3B

Qwen3 Next 80B A3B - Qwen3‑Next 80B mixture‑of‑experts model

Qwen3 Next 80B A3B is the 80‑billion‑parameter Mixture‑of‑Experts member of the Qwen3‑Next family designed to activate a small subset of experts per inference for efficiency. The model emphasizes sparse activation and hybrid attention to deliver long‑context reasoning with lower active compute per query.

Key technical focus

• High‑sparsity MoE lets the model keep an 80B parameter capacity while activating only ~3B parameters per request for improved throughput.
• Hybrid attention / gated mechanisms are used to maintain stability and scaling efficiency on ultra‑long contexts.

Intended usage

The model is intended for production scenarios that need large‑model reasoning with reduced inference cost, especially tasks requiring sustained long‑context understanding and retrieval‑augmented workflows.