LongCat-Flash-Lite-Sparse
Meituan LongCat🇨🇳 CNMajor lab· Jul 31, 2026
Textopen weights
LongCat-Flash-Lite-Sparse is a non-thinking Mixture-of-Experts (MoE) model with 69B total parameters and approximately 3B activated parameters per token. Built on LongCat-Flash-Lite, it replaces dense MLA with LongCat Sparse Attention (LSA) and natively supports context lengths of up to 1M tokens.
From the model card
- License
- mit
- Modalities
- text → text
1 variant
| Standard | HF |
Sources
HuggingFacePaperBenchmarks
| Benchmark | Domain | Value | Evidence |
|---|---|---|---|
| HMMTvendor number | math | 40.53 % correct | 61 |
| AIMEvendor number | math | 65.73 % correct | 53 |
| GPQA Diamondvendor number | reasoning | 66.78 % correct | 50 |
| MMLUvendor number | knowledge | 85.52 % correct | 43 |
| MATH-500vendor number | math | 96.8 % correct | 35 |
Vendor-reported numbers are marked and count with factor 0.7 in rankings. The evidence score says how much a benchmark still tells you today.
Provenance
- Detected
- Sep 27, 2026
- First source
- huggingface
- Description
- From the model card
- Editorially reviewed
- —