LongCat Sparse model released

·2026.08.02 00:10

Key point

LongCat-Flash-Lite-Sparse weights have been released, supporting up to 1M tokens of context.

The weights for LongCat-Flash-Lite-Sparse have been added to the repository and are now available for download.

It is based on the existing LongCat-Flash-Lite, with the following key changes:

Replaces Dense MLA with LongCat Sparse Attention (LSA)
Natively supports up to 1M tokens of context length
Expanded from the existing LongCat-Flash-Lite's maximum context length of 256k tokens

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.