GitHub Repository
Official source code, model card and usage examples for LongCat 2.0.
β Meituan's open-source 1.6T MoE model
Get LongCat 2.0 β Download, API & Docs
Everything you need to download weights, read the docs, call the API, and deploy LongCat 2.0. Full, INT8 and FP8 checkpoints available.
Resources
Official LongCat 2.0 repositories, quantized checkpoints, API platform and related research papers.
Hugging Face Full
Full-precision model weights and tokenizer for research and production.
Hugging Face FP8
8-bit floating-point checkpoint for faster inference on supported hardware.
Hugging Face INT8
8-bit integer quantized checkpoint to reduce memory and latency.
Official API Platform
Access LongCat 2.0 through the official longcat.chat API platform.
LongCat-Flash Tech Report
arXiv report on LongCat-Flash, the efficient inference architecture.
DORA Paper
Research paper describing the DORA method related to LongCat training.
ModelScope
Domestic mirror of LongCat 2.0 weights for faster downloads in China.
Official Launch Blog
Meituan's official LongCat 2.0 announcement and technical deep dive.
API Documentation
Complete English API reference for LongCat 2.0. OpenAI and Anthropic compatible.
API Product Console
Official API onboarding console for integrating LongCat into your app.
OpenRouter
Access LongCat 2.0 through the OpenRouter model hub (previously Owl Alpha).
NPU Inference Code
SGLang-FluentLLM branch with domestic NPU / AI accelerator support.
Model Specs
Key numbers for LongCat 2.0, Meituan's domestically-trainable Mixture-of-Experts language model.
1.6T
Total parameters
33β56B
Active parameters
1M
Context window
MIT
Open license
MoE
Architecture
Domestic
Chip friendly
LongCat Ecosystem
Other models and research releases in the LongCat family.
How to Use
Three simple paths to get started with LongCat 2.0.
Download weights
Grab the full, FP8 or INT8 checkpoint from Hugging Face. Use git clone or the Hugging Face Hub Python library to pull the files.
Read docs / API
Follow the README and examples in the GitHub repo, or sign up at longcat.chat for managed API access.
Deploy or try online
Run inference locally with vLLM/SGLang, self-host on your own GPUs, or skip setup and try the model at trylongcat.com.
FAQ
Common questions about downloading, running and affiliations.
Full-precision weights are available on Hugging Face. Quantized INT8 and FP8 variants are also published there for easier local deployment.
Yes. The official platform and API are hosted at longcat.chat. You can also try the model online at trylongcat.com.
Local inference requires multiple high-memory GPUs for the full model. The INT8/FP8 quantized versions reduce memory footprint, but you should still plan for server-class NVIDIA or domestic accelerators with hundreds of gigabytes of VRAM.
getlongcat.com and trylongcat.com are independent community sites maintained by the same team. getlongcat.com focuses on resource aggregation (downloads, docs, papers), while trylongcat.com provides an online demo for trying the model. Neither is officially affiliated with Meituan.
No. getlongcat.com is a community-maintained resource directory and is not officially affiliated with Meituan. All official releases are published by the
meituan-longcat organization on GitHub and Hugging Face.LongCat 2.0 is released under the MIT license, which permits commercial and research use with minimal restrictions.