Chinese AI company Zhipu claims its new model is a better bug-finder than Anthropic, OpenAI

Chinese AI company Zhipu recently unveiled a new AI model named GLM-5.3, asserting that it rivals the bug-finding abilities of American models from Anthropic and OpenAI. Zhipu's data indicates GLM-5.3 outperforms Fable 5 and GPT-5.6 Sol on the CyberGym benchmark, a test that evaluates a model's expertise in addressing cybersecurity challenges.

According to Zhipu, post-training scaling led to a faster development of cyber capabilities, with GLM-5.3 proving state-of-the-art in vulnerability discovery and excelling in reasoning across multiple stages of exploitation. The company further reported discovering 2,436 vulnerabilities across 269 projects, encompassing medium-to-high severity issues.

Examples include system kernels, operating systems, browser engines, open-source infrastructure, web applications, and network protocols.

While GLM-5.3 excels in bug finding, it underperforms other security and coding benchmarks when compared to Western models. Nevertheless, the fact that China swiftly developed this capability after the debut of Anthropic's Mythos demonstrates that China is not far behind in software evaluation. This development may have diminished the United States' previous advantage in the field of software evaluation.