New Chinese AI model is as good as Anthropic Mythos, report

China is putting more pressure on US AI companies. Researchers say Zhipu AI's latest model, GLM-5.2, can match Anthropic's Mythos in certain bug-finding tasks and even bested OpenAI's GPT-5.5 on select agentic tool-use benchmarks.

In Short

• Chinese AI model GLM-5.2 said to match capabilities of Anthropic's Mythos
• The model is said to be better in certain bug-finding and cybersecurity benchmark tests
• GLM-5.2 is a an open-weight access that gives more users control

China has once again shaken up Silicon Valley's AI ambitions. While Washington and Beijing remain locked in a cold war for artificial intelligence dominance, US companies have long taken pride in staying ahead with models such as Anthropic's Mythos. But that lead may be narrowing. Researchers claim that Chinese AI startup Zhipu AI's latest open-weight model, GLM-5.2, can match Anthropic's flagship Mythos model in certain bug-finding tasks.

GLM-5.2, released earlier this month by Zhipu AI, also known as Z.ai, has reportedly demonstrated performance comparable to some of the best US models when identifying software vulnerabilities. While it still trails leading models from Anthropic and OpenAI in broader reasoning and general AI capabilities, researchers say the gap has narrowed significantly in cybersecurity applications, report WSJ.

The claim for the GLM-5.2 comes at a time when businesses are increasingly looking for lower-cost AI alternatives. Companies including Microsoft are evaluating whether to make Chinese AI models available on their platforms. If more businesses shift towards the low cost models, it could put further pressure on US AI companies.

"China is making sure that the gap becomes smaller and smaller over time," Lior Div, chief executive of cybersecurity company 7AI, told WSJ.

According to researchers, the growing ability of AI models to automatically detect software flaws has increased pressure on organisations to use AI not only to discover vulnerabilities but also to fix them before attackers can exploit them. Some experts have even warned of a future "bugmageddon" if defensive tools fail to keep pace with increasingly capable AI-powered hacking systems.

One of GLM-5.2's biggest differences from models developed by Anthropic or OpenAI is that it is an open-weight model. This means users can download, modify and run it on their own hardware without relying on cloud access. While that gives developers and enterprises greater flexibility, researchers also warn it could make it easier for malicious actors to deploy such systems without oversight.

GLM-5.2 is matching ability of US based AI models

Notably, OpenRouter ranks GLM-5.2 among the world's ten most-used AI models. Cybersecurity company Semgrep also found that, in some benchmark tests, the Chinese model outperformed Anthropic's Claude Opus 4.8. Researchers added that, with additional prompting, both GLM-5.2 and Claude Opus 4.8 were capable of matching Mythos in bug-finding performance.

These developments from Beijing come just as Washington tightens oversight of its most advanced AI systems. Recent restrictions on Anthropic's latest models, combined with closer scrutiny of OpenAI's newest releases, are raising questions about whether stricter U.S. regulations might inadvertently push businesses toward increasingly capable and less restricted Chinese alternatives.

For now, US companies still lead the AI race overall. But China's rapid progress in cybersecurity-focused models suggests that lead is no longer as comfortable as it once was. If this pace continues, Chinese AI will pose a much bigger challenge to America's dominance.

- Ends