DownloadLongCat-2.0 is Meituan’s flagship open-weight Mixture-of-Experts language model designed for frontier-scale coding, reasoning, and autonomous agent workflows. It features 1.6 trillion total parameters with approximately 48 billion activated per token, combining high capability with efficient sparse inference. The model was pretrained on more than 35 trillion tokens and trained entirely on a large-scale cluster of domestically developed AI accelerators, demonstrating stable frontier-scale training without rollback events. LongCat-2.0 introduces LongCat Sparse Attention and extensive 1M-context training, enabling native processing of million-token inputs for long-document analysis, repository-scale coding, and complex multi-step reasoning. Dedicated post-training further strengthens coding and agent performance, producing competitive benchmark results against leading proprietary models.
Features
- 1.6T-parameter Mixture-of-Experts architecture
- Approximately 48B active parameters per token
- Native support for 1M-token context windows
- LongCat Sparse Attention for efficient long-context processing
- Pretrained on more than 35 trillion tokens
- Optimized for agentic coding and software engineering
- Strong benchmark performance in reasoning and coding tasks
- Trained entirely on a large-scale domestic AI accelerator cluster
Project Samples
