ModelsAug 15, 2026 · 12:358 sources

Z.ai Opens GLM-5.3 to Outside Researchers to Validate 84.5 CyberGym Score

From huggingnews.com

AI;DR

Z.ai is offering cybersecurity organizations and independent researchers early access to test the GLM-5.3 model for potential vulnerabilities. The model recently posted a score of 84.5 on CyberGym. Interested parties must complete an application form to participate in the security testing and evaluation.

Read the sourceSign in to suggest

Sources

+ 6 more
  1. SUPPORT@zai_org“invite cybersecurity organizations and security researchers to collaborate with us on evaluating GLM-5.3”x.com
  2. SUPPORT@zixuanli_“It found a potentially serious vulnerability in Cursor”x.com
  3. SUPPORT@arena“post-training on the 743B base model”x.com
  4. SUPPORT@mtslive“found 2,436 vulnerabilities across 269 projects, 1,097 critical or high”x.com
  5. SUPPORT@semianalysis_“massively beats every American open model: Nemotron, Laguna, Inkling”x.com
  6. SUPPORT@genaispotlight“deployed directly on Hugging Face Spaces and GitHub under Apache 2.0”x.com

One ranked edition, every morning

The stories that matter today. Free, cancel anytime.

Telegram @aidr_today