What is ChatLaw?
The official ChatLaw repository describes the project as research on a legal assistant that integrates knowledge graphs, hybrid expert models, and multi-agent systems. The current README lists ChatLaw2-MoE, built on the InternLM architecture and featuring a 4×7B hybrid expert design, for Chinese legal language processing.
The repository simultaneously distinguishes between the earlier ChatLaw-13B and 33B demonstration models and the ChatLaw-Text2Vec text similarity model; the foundations, parameters, and use cases of models from different generations should not be conflated into a single unified specification. The collaborative workflow described in the paper is intended to illustrate the research approach for information collection, retrieval, and report organization.
The primary access points are the repository, the papers page, and the models page; this is not a verified, general-purpose online legal service. The repository displays an AGPL-3.0 license, and the usage requirements for the models and the underlying framework must still be reviewed separately; moreover, publicly available research results do not constitute a guarantee of reliability in specific legal cases.
Key features
Chinese Legal Language Research Model
Knowledge Graph Enhancement
Hybrid Expert Architecture
Multi-Agent Collaboration Research
Legal Text Similarity Model
Public Paper and Model Access
Use cases and audience
Legal tech researchers, model developers, and technology assessment teams
- Legal Language Model Research
- Retrieval and Multi-Agent Process Experiment
- Compare different model generations with the baseline
Legal texts, research data, queries, and model execution configurations
Model outputs, similarity retrieval, or research evaluation results
How to use
- 01
Refer to the official repository to confirm the target model version and its corresponding base.
- 02
Check the runtime requirements and relevant licenses according to the model documentation.
- 03
Conduct controlled evaluations using authorized data, verifying the quality of citations and actual outputs.
Official website & sources
Updated on 2026-10-25
