Zhipu's new legal AI fully passes about one in five law firm tasks.
Chinese lab Zhipu announced GLM-Legal, an AI model trained for legal work, on 12 October. It fully passed 19.2 percent of tasks on Harvey LAB, a test drawn from real law firm work. Testing firm Vals ran that evaluation, and the score is 12.5 points above the base model's.
Why it mattersEven a model tuned for law failed four of five realistic tasks under strict grading. Treat any AI contract review as a draft for a lawyer to check.
Read in the app — free, in 9 languages
Related stories
Meta's first frontier model ranks fourth on an independent index.2026-10-13OpenAI's price list shows a science model it never announced.
2026-10-11Odyssey opened a world model you can steer from a browser.
2026-10-11Google's new image model makes pictures for half the price.
2026-10-10Microsoft's new model answers only with a number, never a sentence.
2026-10-10