Meituan Open-Sources LongCat-Flash-Lite-Sparse with 1M Token Context

LLMFan46 · reddit · 2026-08-01

Meituan has released the weights for LongCat-Flash-Lite-Sparse. Built upon LongCat-Flash-Lite, this model replaces dense MLA with LongCat Sparse Attention (LSA) and natively scales the context length from 256k up to 1 million tokens.

Related event: Meituan Open-Sources LongCat-Flash-Lite-Sparse Model(4 posts)→

Original post →

More from Models

Models channel →