Meituan Open-Sources LongCat-Flash-Lite-Sparse Model

Meituan open-sourced LongCat-Flash-Lite-Sparse, topping Hugging Face trends. Using an ultra-sparse MoE architecture and sparse attention, it activates only 3B of 69B parameters to process 1 million tokens on 24GB VRAM.

2026-07-31 ~ 2026-08-02 · 4 related posts

• Meituan Open-Sources LongCat-Flash-Lite-Sparse: 256k Context on a 24GB GPU — Gohab2001 · 2026-07-31
• Meituan Open-Sources LongCat-Flash-Lite: 69B Total Params with 3B Active — teortaxesTex · 2026-08-01
• Meituan Open-Sources LongCat-Flash-Lite-Sparse with 1M Token Context — LLMFan46 · 2026-08-01
• Meituan Open-Sources LongCat-Flash-Lite-Sparse, Tops Hugging Face Trending — meituan-longcat · 2026-08-02