AI News Feed
Market watch
SSPAI Matrix Dispatch Spotlights Xiaomi Watch S5, EVA Perfume and 512GB Storage DebateMitel WX Targets Frontline, Non-Deskbound Workers With Role-Based UC ClientBose Launches $99 Noise-Cancelling Wired Earbuds as U.S. Presale Sells OutAddigy Announces Intelligence Suite for AI Control on Apple DevicesGooglebook Runs Key Google Services as Web Apps, Not Native Android AppsTFG London investigates cyber-attack that may have exposed customer data at Phase Eight, Hobbs and Whistles
SSPAI Matrix Dispatch Spotlights Xiaomi Watch S5, EVA Perfume and 512GB Storage DebateMitel WX Targets Frontline, Non-Deskbound Workers With Role-Based UC ClientBose Launches $99 Noise-Cancelling Wired Earbuds as U.S. Presale Sells OutAddigy Announces Intelligence Suite for AI Control on Apple DevicesGooglebook Runs Key Google Services as Web Apps, Not Native Android AppsTFG London investigates cyber-attack that may have exposed customer data at Phase Eight, Hobbs and Whistles
Companies

Z.ai Releases GLM-Legal, Scoring 19.17% on Harvey LAB and Beating Claude Fable 5, Kimi K3

Z.ai has released GLM-Legal, a legal AI model built on GLM-5.3-Flash. It scores 19.17% on Harvey LAB, ranking No. 7 of 79, but API access is application-only with no published price or weights.

AI News Feed · 2026-10-13 20:12

GLM-Legal targets contract review, evidence analysis and compliance research for law firms, in-house counsel and compliance teams. Z.ai built it on GLM-5.3-Flash, which the company describes as a 320B-total, 18B-active mixture-of-experts model with a 1M-token context window and 128K maximum output. The base model is natively multimodal, reading text, images, video and files and outputting text. Z.ai has not listed a separate parameter count for GLM-Legal.

Training ran in two stages. Mid-training on expanded legal corpora aimed to strengthen reasoning from facts to rules to conclusions. Post-training used reinforcement learning across tens of thousands of legal task environments. The Z.ai team also described an agent harness co-evolving with the model during post-training. Edge cases in training exposed harness bottlenecks, and harness fixes then exposed the model to more realistic constraints, according to the report.

Harvey LAB is an open-source benchmark of 1,250 tasks across 24 practice areas. More than 75,000 expert-written rubric criteria grade the work, and a task passes only if every criterion passes, a method Harvey calls all-pass grading. Vals AI runs the held-out set with two LLM judges and internet disabled. Agents get file tools plus docx, pptx and xlsx skills. On that leaderboard, GLM-Legal scores 19.17% and ranks seventh of 79. Its base, GLM-5.3-Flash, scores 6.67%. The 12.5-point lift is the core claim, and an independent evaluator reproduced it, the report said. Z.ai's launch chart shows GLM-Legal ahead of Kimi K3 at 12.92%, Claude Fable 5 at 11.25% and GPT-6 Astra at 5.42%. Those figures match Vals AI. The chart omits six models ranked higher, led by Muse Spark 1.2 at 25.42%. GLM-Legal also passes 92.5% of individual criteria, while top models sit near 90% to 95% on that measure.

Z.ai claims parity on LegalBench and the Chinese-focused PLawBench without publishing numbers. Vals AI scores GLM-Legal at 86.23% on LegalBench. Claude Fable 5 leads at 88.56%, and Kimi K3 scores 86.21%. On Vals AI's Legal Research Bench, GLM-Legal scores 41.83%, below its own base model's 45.19%. All benchmark scores, costs and durations cited come from Vals AI's runs as of October 12, 2026. Vals ran both GLM models at $0.075 per 1 million input tokens and $0.25 per 1 million output tokens.

On Harvey LAB, GLM-Legal costs $0.15 per test, compared with $19.23 for Claude Fable 5 and $0.57 for its base model. Its average Harvey LAB duration is 1 hour 12 minutes, longer than the base model's 52 minutes, Kimi K3's 13 minutes 58 seconds, Claude Fable 5's 26 minutes 53 seconds and GPT-6 Astra's 24 minutes 49 seconds. Z.ai has not published an API price for GLM-Legal. The base model is available today at $0.15 per 1 million input tokens and $0.50 per 1 million output tokens, and teams can prototype pipelines on it before swapping the model ID later. The base model self-hosts on multi-GPU servers via vLLM or SGLang with FP8 weights.

Law firms, legal-tech companies and in-house teams can apply for API access to GLM-Legal. No GLM-Legal repository existed on Hugging Face as of October 13, 2026, and Z.ai gave no price. Z.ai says GLM-Legal will come to AutoClaw, its desktop agent app, but gave no date.