LongCat-2.0 is an MIT-licensed MoE language model from Meituan LongCat with 1.6T total parameters, about 48B active parameters, and a 1M-token context window. It uses LongCat Sparse Attention and is adapted for coding, repository-level understanding, tool use, and long-horizon agent workflows.
Features
- 1.6T total MoE parameters
- About 48B active parameters
- 1M context window
- LongCat Sparse Attention
- MIT open weights
- Post-trained for coding and agents
Use Cases
- Long-context code understanding
- Repository-level Q&A
- Agent workflow model evaluation
- Open model research
- Tool-use experiments
- Local or private model deployment evaluation
FAQ
LongCat-2.0 is an MIT-licensed MoE language model from Meituan LongCat with 1.6T total parameters, about 48B active parameters, and a 1M-token context window. It uses LongCat Sparse Attention and is adapted for coding, repository-level understanding, tool use, and long-horizon agent workflows. Core capabilities include: 1.6T total MoE parameters, About 48B active parameters, 1M context window.
Common scenarios include: Long-context code understanding, Repository-level Q&A, Agent workflow model evaluation.
