China’s Zhipu AI, also known as Z.ai, has released the open-weight GLM-5.2 model, which researchers say can match Anthropic’s Mythos in some bug-finding and cybersecurity scenarios.
Since it’s an open-weight model, it is free for anyone to download and use as they please.
The model’s performance shows that Chinese AI developers are narrowing the capability gap with leading US companies.
However, its release has also raised security concerns because anyone can download, run, and modify the model with limited oversight.
Open-Weight Risks
GLM-5.2’s open-weight release allows users to operate it on readily available hardware and access its underlying parameters.
This gives developers greater control and allows organisations to run the model within their own systems.
However, malicious users could also modify it or use it to search for software vulnerabilities without the restrictions applied to closed models.
The US government has tried to restrict China’s access to advanced systems such as Anthropic’s Mythos and Fable because of concerns about their cybersecurity capabilities.
OpenAI’s GPT-5.6 has prompted similar concerns over potential misuse and is available under limited access restrictions.
Model Capabilities
GLM-5.2 has 753 billion parameters and a context window of one million tokens.
OpenRouter has ranked it among the ten most-used AI models worldwide.
The model supports adjustable reasoning levels, allowing users to choose between quicker responses and more detailed processing depending on the task.
Its large context window also allows it to work through long coding sessions and large software repositories without losing earlier information.
Cybersecurity company Semgrep tested GLM-5.2 on software vulnerability detection and found that it outperformed Claude Opus 4.8 in some scenarios.
The result applied to a specific test and did not show that GLM-5.2 performs better across every cybersecurity task.
Coding Performance
GLM-5.2 scored 62.1 on SWE-bench Pro, a benchmark that measures performance on software engineering tasks.
That placed it ahead of GPT-5.5, which scored 58.6, but behind Claude Opus 4.8 at 69.2.
Its open-weight format allows users to download and change the model for use without external supervision.
While this benefits developers who need greater control, hackers could also use it to identify security weaknesses without safeguards or monitoring.
Another Chinese Tool
Chinese cybersecurity company 360 Security Technology recently introduced a fuzzing tool called Tulongfeng.
The company says the tool has bug-finding capabilities similar to Anthropic’s Mythos.
“This kind of powerful weapon that can alter the landscape of cyberwarfare can’t remain solely in American hands,” 360 Security Chief Executive Zhou Hongyi said.
Stay Connected with ProPakistani
Get the latest tech news, telecom insights, and product launches wherever you prefer.
Add ProPakistani to Preferred Sources and see more of our stories in Google Search and Top Stories.
