AI news ·
Lower costs and customization drive increased use of Chinese open-weight AI models in legal tech
Chinese open-weight AI models cost one-fifth the price of U.S. alternatives for legal inference. Smaller firms adopt them to keep sensitive client data on local servers.
Chinese open-weight AI models are reshaping the economics of legal technology, offering law firms and vendors a path to lower-cost, customizable tools that can be fine-tuned on proprietary data. The shift comes as firms grapple with tight budgets and a growing appetite for generative AI, while U.S.-based models from OpenAI and Anthropic continue to dominate the high-end market.
The rise of open-weight models in legal tech
Open-weight models, unlike fully closed systems, allow developers to download, modify, and run the model on their own infrastructure. This gives legal tech vendors greater control over data security and the ability to tailor the model for specific legal tasks-contract review, e-discovery, or legal research-without paying per-token fees. Deepseek and other Chinese models have gained traction among vendors who serve small to midsize law firms that cannot absorb the cost of large-scale commercial API access.
Several legal technology companies are now testing these models for document summarization and clause extraction. Early results show that for narrowly defined tasks, the performance gap with U.S. models narrows considerably. The ability to fine-tune on a firm's own contract language or compliance documents makes the output more predictable, a key requirement for legal work.
U.S. models still lead on performance, but at a cost
OpenAI and Anthropic continue to hold an edge in raw performance on complex reasoning and multi-step analysis. Their models handle nuanced legal arguments and regulatory interpretation with fewer errors, according to internal benchmarks shared by several vendors. However, the cost differential remains stark: one legal tech executive said their firm's monthly inference bill for a U.S. model ran five times higher than what they now spend running an open-weight alternative on dedicated servers.
That price gap is driving a split in the market. Large firms with deep pockets still use U.S. models for high-stakes work, while smaller firms and specialized boutiques adopt open-weight systems for routine drafting and review. AI for legal applications is increasingly defined by this tiered approach.
Privacy and sovereignty concerns drive adoption
For firms handling sensitive client data, the ability to run models on-premises is a critical advantage. Many partnership agreements and corporate clients now require that AI tools not send data to external servers. Open-weight models let firms keep everything in-house, a feature that Chinese models have leaned into, as they are often accompanied by fewer usage restrictions and less aggressive data-collection policies than some U.S. counterparts.
Regulatory pressure is also a factor. In jurisdictions with strict data sovereignty laws, the option to deploy a model that never leaves the firm's own data center removes a major compliance hurdle. Vendors that once hesitated to build on non-U.S. models now see them as a practical necessity for certain client segments.
Why this matters for legal professionals
For practicing lawyers and legal operations teams, the proliferation of open-weight models means more choice in the AI tools they can deploy. Firms can now evaluate AI based on cost, data privacy, and task-specific accuracy rather than defaulting to a single, expensive provider. The shift does not eliminate the need for human review-no model is error-free-but it does lower the barrier to entry for AI-assisted legal work. In-house teams should start testing open-weight models on non-critical tasks to build institutional knowledge before the technology becomes standard in their practice area.