SignalCapabilitySG-0016
Zhipu's GLM-5.3 matched US frontier models at finding bugs, with open weights to follow
Chinese developer Zhipu launched GLM-5.3, a coding model it says developed unexpectedly strong cyber skills. It scored slightly above Mythos 5 and GPT-5.6 Sol on the CyberGym vulnerability benchmark, though well behind both at building working exploits. Zhipu said expert review confirmed 2,436 vulnerabilities across 269 projects, about 1,100 of them medium to high severity, and planned to publish the model's weights roughly two weeks after launch, after safety evaluation.
Why it matters
Publishing the weights would let others run and modify the model without relying on the developer's hosted service. That could make access restrictions and monitoring harder to enforce.
Sources
Read the reporting
3 sources. Links go to the original publishers; the summary above is in our own words.
- primaryGLM-5.3: Frontier Coding with Emergent Cyber CapabilitiesZ.ai · Aug. 14, 2026z.ai/blog/glm-5.3
- newsZhipu says new coding AI developed advanced cyber skills faster than expectedCSO Online · Aug. 17, 2026csoonline.com/article/4210501
- newsZ.ai says GLM-5.3 nears Anthropic in cyber defense as coding gains accelerateKrASIA · August 2026kr-asia.com/z-ai-says-glm-5-3-nears-anthropic-in-cyber-defense-as-coding-gains…
Signals
More from the news desk
Cite and share
Use this signal
Citation
Paperclip Index. “Zhipu's GLM-5.3 matched US frontier models at finding bugs, with open weights to follow.” Signal SG-0016. Reported Aug. 14, 2026; updated Oct. 7, 2026. https://paperclipindex.com/signal/SG-0016