LongCat-Flash-Chat: 560B MoE Model with 27B Active Params Sets New Benchmarks
LongCat-Flash-Chat, an open‑source 560‑billion‑parameter Mixture‑of‑Experts model that activates only 18.6‑31.3 B parameters per token, delivers state‑of‑the‑art performance on general, agentic, coding, and instruction‑following benchmarks while offering fast inference and efficient deployment options.
