Zhipu AI's GLM-5.2 matches Anthropic's cybersecurity benchmarks at a fraction of the cost, raising uncomfortable questions about US export controls
China’s AI labs have been playing catch-up with their American counterparts for years. That gap just got a lot smaller, and it happened in one of the most sensitive domains imaginable: cybersecurity.
Zhipu AI, the Chinese company also known as Z.ai, released its open-weight GLM-5.2 model in June 2026. The model matches or exceeds Anthropic’s Mythos in cybersecurity benchmarks, including automated vulnerability detection and static code analysis. It costs about one quarter the price per token.
The price-performance equation changes everything #
At approximately 25% of the per-token cost of comparable US systems, GLM-5.2 doesn’t just compete on performance. It competes on economics.
Venture capitalist Marc Andreessen characterized GLM-5.2 as the first Chinese model that can consistently match or surpass American models without compromise, calling it a new standard for Chinese AI.
Jefferies strategist Christopher Wood described GLM-5.2 as “almost equal” to Anthropic’s model in the corporate market, with particular emphasis on its cost efficiency.
GLM-5.2 still lags behind US models in broader application capabilities. The model excels in targeted cyber tasks but doesn’t match the all-around versatility of leading American systems.
The export control paradox #
Georgetown researcher Sam Bresnick called the advancements in Chinese AI a significant “wake-up call” for cybersecurity development.
The concern among security analysts isn’t just about commercial competition. When a model that excels at finding software vulnerabilities is open-weight, meaning its architecture is publicly available for modification, the implications extend well beyond corporate cybersecurity budgets. The same capabilities that help defenders find bugs can help attackers exploit them.
What this means for investors #
American firms that have built their competitive moat on having the best AI-powered security tools just watched that moat get narrower. If a Chinese open-weight model can match their performance at one quarter the cost, pricing power erodes.
On the other side, cheaper cybersecurity AI could expand the total addressable market. A model at 25% of the token cost suddenly makes those capabilities accessible to organizations that were previously running on legacy defenses.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our