Chinese AI company Z.ai, known in China as Zhipu AI and the creator of the GLM model family, has launched GLM-Legal, an LLM it describes as 'specifically trained and optimised for legal work.' The model is built on Z.ai's existing GLM-5.3-Flash model with additional legal-domain training and reinforcement learning, and is designed to handle tasks ranging from contract review to regulatory compliance. It is multi-lingual.

The company said it also plans to bring its legal capabilities into AutoClaw, its desktop agent that handles tasks such as spreadsheet analysis and business operations. Although Z.ai trained GLM 5.3 Flash, the article understands the legal version is not intended, at least for now, to be an open weights model. Z.ai ran GLM-Legal through several legal AI benchmarks, including Harvey's, with results shown in data and tables on the company's page.

The article notes that major AI players have entered legal mainly by encouraging legal tech companies to connect to their models or by building skills and add-ins for lawyers, rather than post-training their core LLMs for lawyers. Thomson Reuters has used open weights models to build its own legal LLM, Thomson, from its research data, and some law firms are building smaller models or working with legal tech companies on narrow solutions. GLM-Legal differs because it is a standalone legal-specific model from an LLM maker that can be used as is, rather than an add-in.

Whether law firms will adopt it is unknown. The article says lawyers will want more information on where data goes, as with any new model, and pricing has not been disclosed. It adds that LLMs from China are generally more affordable than well-known Western models. The author sees the launch as a sign that China is developing cutting-edge legal AI tools, noting a surge of legal tech tools in China a decade ago that faded and now appears to be growing again, with GLM-Legal pitched at the global market rather than only domestic use.

Chinese AI company Z.ai, known in China as Zhipu AI and the creator of the GLM model family, has launched GLM-Legal, an LLM it describes as 'specifically trained and optimised for legal work.' The model is built on Z.ai's existing GLM-5.3-Flash model with additional legal-domain training and reinforcement learning, and is designed to handle tasks ranging from contract review to regulatory compliance. It is multi-lingual.

The company said it also plans to bring its legal capabilities into AutoClaw, its desktop agent that handles tasks such as spreadsheet analysis and business operations. Although Z.ai trained GLM 5.3 Flash, the article understands the legal version is not intended, at least for now, to be an open weights model. Z.ai ran GLM-Legal through several legal AI benchmarks, including Harvey's, with results shown in data and tables on the company's page.

The article notes that major AI players have entered legal mainly by encouraging legal tech companies to connect to their models or by building skills and add-ins for lawyers, rather than post-training their core LLMs for lawyers. Thomson Reuters has used open weights models to build its own legal LLM, Thomson, from its research data, and some law firms are building smaller models or working with legal tech companies on narrow solutions. GLM-Legal differs because it is a standalone legal-specific model from an LLM maker that can be used as is, rather than an add-in.

Whether law firms will adopt it is unknown. The article says lawyers will want more information on where data goes, as with any new model, and pricing has not been disclosed. It adds that LLMs from China are generally more affordable than well-known Western models. The author sees the launch as a sign that China is developing cutting-edge legal AI tools, noting a surge of legal tech tools in China a decade ago that faded and now appears to be growing again, with GLM-Legal pitched at the global market rather than only domestic use.