Z.ai says GLM-5.3 scores 84.5% on CyberGym, vs. Mythos 5's 83.8%, and its most sensitive cybersecurity functions will only be available to verified users

I'm coming down from spending a few days at Usenix Security …Tomoko Wakasugi /Nikkei Asia:China's Z.ai launches model it says rivals Anthropic's MythosSam Sabin /Axios:A Chinese lab's new model is nearly as good at hacking as U.S. AIDylan Kresak /The Daily Caller:CCP-Linked AI Rivals US Anthropic's Model In Cyber TestsSouth China Morning Post:Zhipu launches flagship model GLM-5.3 as China seeks Mythos-level edge in cyber defenceZ.ai:How to Switch ModelsTom Krazit /The Stack:For Z.ai's GLM-5.3, p

Reuters 2026-08-14

Context & Ripple Effects

Z.ai’s GLM releases have moved quickly from an MIT-licensed GLM-5.1 framed around software-engineering benchmarks to GLM-5.2, which researchers said matched leading U.S. models at finding security bugs. GLM-5.3 extends that security-focused comparison to CyberGym while changing access for its most sensitive functions.

The access restriction is notable against related coverage’s debate over restrictions on Chinese open models: Z.ai is pairing a frontier-performance claim with verified-user gating rather than treating broad availability as the only distribution model.

First-order effects

Verified users, rather than the full model audience, will receive access to GLM-5.3’s most sensitive cybersecurity functions.
Z.ai puts an explicit CyberGym comparison with Anthropic’s Mythos 5 at the center of GLM-5.3’s launch, claiming an 84.5% score versus 83.8%.

Second-order effects

The verified-user requirement creates a separate access tier for cybersecurity use, departing from the broad MIT-license availability associated with GLM-5.1.
Anthropic and other frontier-model providers face a more visible benchmark comparison in cyber capabilities, while users must weigh claimed performance against the provider’s access controls.

Third-order effects

If leading labs continue coupling cyber-capability gains with user verification, access policy may become a core competitive dimension alongside benchmark results.
The pattern supports a shift toward cyber-capable models being treated as strategically controlled infrastructure, especially amid the debate documented around restrictions on Chinese open models.

The trend: Frontier AI labs are increasingly pairing cybersecurity performance claims with controlled access, making model governance part of the product race.

Discussion

@samhogan Sam Hogan on x
GLM 5.3 benchmarks look great, pretty substantial improvement over 5.2 on coding and cyber
@hesamation @hesamation on x
While Anthropic and OpenAI put cyber defense behind safeguards, GLM 5.3 is specifically trained for coding and cyber defense. Wow.
@elshayib_ @elshayib_ on x
GLM-5.3 Sucks come on @Zai_org you're doing it wrong, you can't say your model is good for cyber defense with 0 Sandbox escapes, come on step up your game.
@zixuanli_ Zixuan Li on x
Building on GLM-5.2's contributions to cyber defense, GLM-5.3 delivers substantially stronger capabilities in vulnerability discovery, exploit analysis, and complex, multistep security tasks in realistic environments. These advances can help defenders identify weaknesses earlier,
@joshua_saxe Joshua Saxe on x
Really cool to see a https://z.ai/ / GLM engineer demonstrating defensive cyber capabilities of their new model. Feels like there's a real possibility that if the cyber guardrails on the American closed models don't change the security community will move over to these
@zaddyzaddy @zaddyzaddy on x
The cybersecurity doom-mongering from the big labs is about to lose its force. GLM-5.3 has shown that existing models can become significantly more capable on cyber tasks with additional RL training. So what stops others from building on GLM-5.2 and pushing its cybersecurity
@louszbd Lou on x
We gave GLM-5.3 a complex reverse-engineering task. It found a potentially serious vulnerability in Cursor. We disclosed it privately. Appreciate Cursor team is working closely with us on a fix, and we'll share the more details once users are protected.
@_nathancalvin Nathan Calvin on x
“API access and open weights will be released in stages following rigorous safety evaluations.” GLM 5.3 seems to be extremely good at cyber defense and cyber offense, curious what rigorous safety evaluations entails

Chronicles