LongCat Flash Lite needs about 41.2 GB of memory at Q4_K_M (a 40.2 GB file) with an 8K context window, so it fits on a 48 GB GPU. This repository has 6 quants, from 32.8 GB (Q3_K_L) to 137.1 GB (BF16). Longer context needs more: at 32K tokens the KV cache alone adds about 2.0 GB.
68.5B parameter Mixture-of-Experts model with 327,680 token context
LongCat Flash Lite is a 68.5B parameter MoE language model. It activates only 3–4.5B parameters per token and supports a 327,680 token context window. The model uses N‑gram embeddings, Multi‑head Latent Attention, and identity experts for efficient inference.
Good at
- Agentic tool use
- Coding proficiency
- Long‑context inference
Limits
- Requires custom llama.cpp fork
- No vision support
- Large VRAM needed for full precision
Best for
- Long‑form content generation
- Agentic task automation
- Code generation
Who it’s for. Developers and researchers needing a large‑scale MoE model with efficient long‑context handling. The model is suitable for applications that can leverage the custom architecture and have access to high‑VRAM GPUs.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| Q3_K_L | 32.8 GB | 33.8 GB | Too big | Get file → | |
| Q4_K_M | 40.2 GB | 41.2 GB | Too big | Get file → | |
| Q5_K_M | 48.0 GB | 49.0 GB | Too big | Get file → | |
| Q6_K | 56.3 GB | 57.3 GB | Too big | Get file → | |
| Q8_0 | 72.9 GB | 73.9 GB | Too big | Get file → | |
| BF16 | 137.1 GB | 138.1 GB | Too big | Get file → |
For 8 GB VRAM, Q3_K_L (32.79 GB) is the smallest but may degrade quality. Q4_K_M (40.23 GB) offers a better balance of quality and speed. For 16 GB VRAM, Q5_K_M (48.01 GB) is recommended. Memory estimates are approximate.
What GPU do you need for LongCat Flash Lite?
These figures add the file size, an fp16 KV cache at an 8K context window (worked out from this model’s own header) and about 0.5 GB for the runtime. They are estimates, so leave a little headroom, and expect to use less context or a smaller quant if you are right at the limit.
- 4 GB VRAM: no quant fits at an 8K context.
- 6 GB VRAM: no quant fits at an 8K context.
- 8 GB VRAM: no quant fits at an 8K context.
- 12 GB VRAM: no quant fits at an 8K context.
- 16 GB VRAM: no quant fits at an 8K context.
- 24 GB VRAM: no quant fits at an 8K context.
- 32 GB VRAM: no quant fits at an 8K context.
- 48 GB VRAM: the best quant that fits is Q4_K_M (40.2 GB file, about 41.2 GB in use).
Which LongCat Flash Lite quant should you choose?
Q4_K_M is the usual starting point for LongCat Flash Lite: it balances file size and quality well. Here it is a 40.2 GB file.
The Q8_0 file is 72.9 GB, about 1.8 times the size of Q4_K_M. It keeps the weights closer to the original, which only matters if you have the memory to spare.
If memory is tight, the smallest file is Q3_K_L at 32.8 GB, with the quality trade-offs that come with fewer bits. A reasonable choice when a 4-bit file will not fit, mostly on larger models. Quality loss is noticeable but often acceptable.
New to quant names? Read GGUF quantization explained.
Specifications
Architecture from the GGUF header
Source
Frequently asked questions about LongCat Flash Lite
How much VRAM does LongCat Flash Lite need?
About 41.2 GB at Q4_K_M with an 8K context window. That is the 40.2 GB file, plus the KV cache and roughly 0.5 GB of runtime overhead. Smaller quants need less and longer context needs more.
Can I run LongCat Flash Lite on an 8 GB GPU?
Not at an 8K context. The smallest file needs about 33.8 GB, so you would need a larger GPU or CPU offloading.
What is the file size of LongCat Flash Lite GGUF?
It depends on the quant: Q3_K_L is 32.8 GB, Q4_K_M is 40.2 GB, Q5_K_M is 48.0 GB, Q6_K is 56.3 GB, Q8_0 is 72.9 GB, BF16 is 137.1 GB.
What context length does LongCat Flash Lite support?
The GGUF header lists a trained context length of 327,680 tokens (320K). The KV cache grows with the context you use: about 0.5 GB at 8K and 2.0 GB at 32K.
Where can I download LongCat Flash Lite GGUF?
The files are on Hugging Face in the InquiringMinds-AI/LongCat-Flash-Lite-GGUF repository. The download buttons on this page open the file on Hugging Face. ggufmodels.com does not host any model files.
What license does LongCat Flash Lite use?
Hugging Face lists the license as mit. Licenses can differ between the quantized files and the original model, so check the repository and the base model before commercial use.
Can I run LongCat Flash Lite on 8 GB VRAM?
The smallest quantization is Q3_K_L at 32.79 GB, which exceeds 8 GB VRAM. It cannot run on 8 GB VRAM.
Does this model support vision tasks?
No, LongCat Flash Lite is a text‑only language model.
What fork of llama.cpp is required?
A custom fork from the longcat‑flash‑ngram branch is required to run this model.