appConfigEntityId

bodyType

customContents

entityId

entityUuid

isVideo

articlePageQuery

article

articleSchemaContent

content

hooksArticlePageAdvertisementConfigArticle

hooksArticlePageFloatingButtonConfigArticle

appConfig

filter

json

liveArticle

Xinmei Shen

Richard Chen

Chinese artificial intelligence firm Zhipu, also known as Z.ai, has unveiled its flagship GLM-5.3 model, saying it beat Anthropic’s frontier Mythos 5 model in a key cybersecurity test, as China races to counter Western advances in AI defence.
Beijing-based Zhipu said GLM-5.3 achieved a success rate of 84.5 per cent on CyberGym, a benchmark that measures whether models can identify and validate security flaws from source code.
That was above Anthropic’s Mythos at 83.8 per cent and OpenAI’s GPT-5.6 Sol at 83.6 per cent, according to Zhipu.
However, the Chinese model did not match those foreign systems on ExploitBench, which gauges how far AI models climb the exploitation ladder. Its score of 54.4 per cent trailed Mythos’ 78 per cent and GPT-5.6 Sol’s 76.5 per cent.
AI development should not be a solo performance by one nation, but a symphony of global collaboration
Zhipu
Zhipu said it had tested the model with security teams in China against real-world codebases, identifying 2,436 vulnerabilities across 269 projects after expert review. Of those, 1,097 were rated medium to high severity, according to the company.
It said it planned to release GLM-5.3’s weights – the underlying parameters that encode its intelligence – in about two weeks, after an internal safety review and hardening process. It described the model’s cyber-defensive capabilities as “public goods” that developers worldwide could access and improve.
“AI development should not be a solo performance by one nation, but a symphony of global collaboration,” Zhipu said in its announcement. It added that open-sourcing GLM-5.3 would mean defensive capabilities were no longer “privileges of a few”.
GLM-5.3 uses the same base model as its predecessor GLM-5.2, a 744-billion-parameter model introduced in June.
Zhipu boasted that GLM-5.3’s coding abilities had approached that of Anthropic’s Fable 5, saying it “significantly raised the ceiling of” GLM-5.2’s intelligence through “extreme post-training scaling”.
GLM-5.3 received a score of 28.5 on the famously difficult evaluation Agent’s Last Exam, becoming the best open-weight model on the test, ahead of Kimi K3’s 27.6. GLM-5.3 closely trailed GPT-5.6-Sol, which scored 28.6.
The earlier model, GLM-5.2, was deployed by developer platform Hugging Face to help contain an autonomous cyberattack involving OpenAI’s frontier systems, bringing open-weight models into the debate over whether wider access to advanced AI can also improve software security.
Mythos, released in April, is designed to autonomously identify software vulnerabilities. Anthropic said it accelerated discovery a hundredfold while cutting costs.
At a cybersecurity conference in Beijing in June, 360 Security Technology founder Zhou Hongyi called the advance a “terrifying change” that had effectively “democratised” cyberattacks by lowering barriers to exploitation.
China’s cybersecurity industry needed its own version of Mythos, a “game-changing weapon in cyber warfare” that “cannot be held solely in the hands of others”, Zhou said.
A Jefferies research note on Friday said GLM-5.3’s release, together with DeepSeek-V4, reinforced a growing emphasis on post-training, “as model intelligence is not determined solely by [the number] of parameters”.
Chinese AI labs are racing to expand parameters to train their systems. For example, Moonshot AI’s Kimi K3 and Alibaba Group Holding’s Qwen3.8-Max each exceed 2 trillion parameters.
Jefferies analysts expected “intensifying competition” in the industry with “an increasing emphasis on cost efficiencies”.


Philippine tribe vows to stand ground against US-led Pax Silica: ‘listen to us’

As the US weighs restrictions on cloud computing, how will China’s AI sector adapt?

Chinese AI chipmaker Hygon plots expansion from data centres to robotics

For China’s hi-tech IPOs, it’s Hong Kong vs the mainland – but how much does it matter?

Artificial intelligence (AI)

Zhipu launches flagship model GLM-5.3 as China seeks Mythos-level edge in cyber defence

Zhipu says it has tested GLM-5.3 with security teams in China against real-world codebases, identifying 2,436 vulnerabilities across 269 projects after expert review. Photo: Shutterstock 

Chinese doctor stuns maths world by cracking decades-old problem using ChatGPT

Frontier artificial intelligence models are making rapid progress in mathematics. Photo: Shutterstock

DeepSeek’s updated V4 Pro AI model struggles on benchmarks, shines in cybersecurity

Chinese artificial intelligence start-up DeepSeek quietly released an updated version of its latest flagship model, DeepSeek-V4-Pro-0813, on Wednesday. Photo: Reuters

Meta to challenge China’s open-weight AI dominance amid US regulatory fears

Meta CEO Mark Zuckerberg attends the annual Allen and Company Sun Valley Media and Technology Conference at the Sun Valley Resort in Sun Valley, Idaho on July 9, 2026. Photo: Reuters

Meta Platforms has renewed its push into open-source models in a bid to lure away American customers who are worried about looming regulatory restrictions in Washington due to the adoption of Chinese AI. Photo: Reuters

Alibaba tests paid AI appetite with US$30 annual office-assistant subscription

The e-commerce giant’s Qwen app now offers three membership tiers for the office assistant. Photo: Shutterstock

Tech

Big Tech

Policy

Tech Trends

Tech War

Business

Company says GLM-5.3 outperformed leading US systems in cybersecurity tests, beating Anthropic’s Mythos 5 and OpenAI’s GPT-5.6 on CyberGym


Company says GLM-5.3 outperformed leading US systems in cybersecurity tests, beating Anthropic’s Mythos 5 and OpenAI’s GPT-5.6 on CyberGym.


Zhipu launches flagship model GLM-5.3 as China seeks Mythos-level edge in cyber defence

.css-1c6uqr6{color:inherit;font-weight:inherit;font-size:inherit;font-family:inherit;line-height:inherit;overflow-wrap:break-word;}Company says GLM-5.3 outperformed leading US systems in cybersecurity tests, beating Anthropic’s Mythos 5 and OpenAI’s GPT-5.6 on CyberGym

Company says GLM-5.3 outperformed leading US systems in cybersecurity tests, beating Anthropic’s Mythos 5 and OpenAI’s GPT-5.6 on CyberGym