China’s Z.ai claims it can match Mythos on cybersecurity
TL;DR
Zhipu AI, now branded as Z.ai, has released the open-weight GLM-5.2. According to The Verge, some researchers say it can match Mythos in specific bug-finding and cybersecurity scenarios. GLM-5.2 still appears to trail top Anthropic and OpenAI models on broader general-purpose tasks. The story is not that China has fully caught up; it is that the gap may be shrinking fast in security-relevant specialist tests.
Nauti's Take
This reads like a classic geopolitics alarm story, but the sober takeaway is more important: specialist capabilities can diffuse faster than overall model leadership. For companies, that does not mean panic; it means tighter code security, dependency scanning, and red-teaming.
Anyone assuming that only a few US labs can deliver serious cyber capability should now treat that assumption as outdated.
Briefingshow
Cybersecurity is not a neutral benchmark category. A model that finds vulnerabilities better can help defenders, but it can also make it easier for attackers to identify exploitable bugs. If those capabilities appear in open-weight models, the debate shifts from who has access to who can operationalize them fastest.
Sources