LongCat-Flash-Lite-Sparse
Meituan LongCat🇨🇳 CNMajor lab· Jul 31, 2026
LongCat-Flash-Lite-Sparse is a non-thinking MoE model with 69B total and about 3B active parameters per token, supporting up to 1M tokens of context. It reports gains over its dense predecessor in agentic coding and tool use, e.g. 68.20 vs 54.40 on SWE-Bench Verified.
Summary of the model card
- License
- mit
- Modalities
- text → text
1 variant
| Standard | HF |
Sources
HuggingFacePaperDeutsch
LongCat-Flash-Lite-Sparse ist ein Mixture-of-Experts-Modell ohne Thinking-Modus mit 69B Parametern, von denen rund 3B pro Token aktiv sind. Laut Modelcard verarbeitet es bis zu 1M Tokens und liegt bei Agentic Coding und Tool-Use vor seinem Vorgaenger, etwa 68.20 gegenueber 54.40 bei SWE-Bench Verified.
Benchmarks
| Benchmark | Domain | Value | Evidence |
|---|---|---|---|
| HMMTvendor number | math | 40.53 % correct | 58 |
| MATH-500vendor number | math | 96.8 % correct | 55 |
| AIMEvendor number | math | 65.73 % correct | 53 |
| GPQA Diamondvendor number | reasoning | 66.78 % correct | 49 |
| MMLUvendor number | knowledge | 85.52 % correct | 39 |
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
- Summary of the model card
- Editorially reviewed
- —