LongCat-Flash-Lite-Sparse
meituan-longcat/LongCat-Flash-Lite-Sparse
Official · BF16 · safetensors
Runtime certified: SGLang
| Precision / quantization | Format | Bits | Publisher trust | Deployment state |
|---|---|---|---|---|
| BF16 | safetensors | 16 | Official | Runtime certified |
LongCat · Flash · LongCat Flash Lite
LongCat Flash Lite Sparse: Specs, GPU requirements, hosting & deployment
LongCat Flash Lite Sparse is a current Flash release in the LongCat family, certified from its immutable official source revision.
Total parameters69.0B
Active parameters3.0B
Context window983,040 tokens
Release statusCurrent · Current
ArchitectureLongcatCausalLM
ModalitiesText
Licensemit
Deployment stateDeployable with certified configuration
Commercial stateCommercial Use Allowed
Release date2026-07-31
Architecture, strengths and use cases
LongCat Flash Lite Sparse is an official self-hostable Meituan release in the LongCat Flash track. It uses LongcatCausalLM, carries 69B / 3B MoE, supports a verified context window of 983,040 tokens, and exposes only source-certified checkpoints and runtime combinations. Its release, license and deployment facts are tied to the official repository revision rather than inferred from the family name.
LongCat Flash Lite Sparse uses LongcatCausalLM with 69B / 3B MoE. The stored architecture profile records 14 layers, hidden width 3072, 32 attention heads and a 983,040 token context ceiling where the official configuration publishes those fields.
Verified LongcatCausalLM architecture
983,040 token certified context
Official BF16, Unknown checkpoint coverage
Private LLM inference
Agentic workflows
Long-context document processing
Certified checkpoints and quantizations
Only exact, source-certified variants attached to this release are shown.
Recommended hardware
Use only the listed sglang runtime configuration with the exact checkpoint precision. The recommendation engine checks weight memory, KV cache, runtime overhead, safety headroom, GPU architecture and topology before ranking rentable and reference hardware.
Estimated weight memory132.38 GB
KV cache0.44 GB
Runtime overhead33.68 GB
Total with safety margin206.45 GB
Recommended best value
2× NVIDIA H200 SXM 141GB
282 GB total VRAM · nvswitch
75.55 GB headroom · Limited capacity
USD 4,690 / monthLowest-cost viable
2× NVIDIA H200 SXM 141GB
282 GB total VRAM · nvswitch
75.55 GB headroom · Limited capacity
USD 4,690 / monthHighest performance
2× NVIDIA H200 SXM 141GB
282 GB total VRAM · nvswitch
75.55 GB headroom · Limited capacity
USD 4,690 / monthLicense and self-hosting considerations
The full 69B resident weight set must fit across the selected GPUs. Context and concurrency add KV-cache memory; MoE active parameters do not replace resident-weight memory.
LongCat Flash Lite Sparse is published under mit. The release-scoped official license link and verification date are retained with the catalogue record.
Commercial Use Allowed. The official license permits commercial use; its obligations still apply.
Release timeline
meituan-longcat/longcat/longcat-flash-lite-sparse
Source details
Source revision: 84a85c4a0529f4faea6ae9cdbef855d40ed557e6afe5f1449b1936c1f4af54d6
Frequently asked questions
How much VRAM does LongCat Flash Lite Sparse require?
The exact estimate depends on checkpoint precision, context and concurrency; the page recommendation uses the certified memory engine and compatible runtime configurations.
Which runtimes support LongCat Flash Lite Sparse?
Certified upstream runtime evidence is available for sglang.
Can LongCat Flash Lite Sparse be used commercially?
The stored commercial classification is Commercial Use Allowed. Consult the official mit terms for the final obligations.
Official references
Last verified: 2026-09-03