Chinese startup claims AI model matches Anthropic's Mythos as US-China AI race heats up

Chinese artificial intelligence company Zhipu AI, also known as Z.ai, has said its latest AI model, GLM-5.2, can match Anthropic's Mythos in certain cybersecurity tasks, according to a Wall Street Journal report.Researchers found that while the Chinese model has reached a similar level in identifying software security bugs, it still trails Anthropic and OpenAI's models on broader AI tasks.Gap with US AI models narrowsThe Wall Street Journal reported that the progress made by GLM-5.2 could reshape the AI race between the US and China, especially as Washington tightens restrictions on access to advanced American AI models.The Trump administration recently asked Anthropic to limit access to its Fable 5 and Mythos 5 models outside the US. OpenAI was also asked to stagger the release of GPT-5.6.What researchers foundAccording to OpenRouter, GLM-5.2 is among the world's 10 most-used AI models.Cybersecurity company Semgrep found that GLM-5.2 outperformed Claude Opus 4.8 in some bug-finding tests. Researchers also said that with additional prompting, both GLM-5.2 and Claude Opus 4.8 were able to match Mythos in identifying software vulnerabilities.On the SWE-bench Pro coding benchmark, GLM-5.2 scored 62.1, ahead of GPT-5.5's 58.6 but behind Claude Opus 4.8's 69.2.Open-weight model raises concernsGLM-5.2 is a 753 billion-parameter open-weight model with a one million-token context window. It also lets users adjust the level of reasoning depending on whether they want faster or more detailed responses.Because the model can be downloaded, modified and run on local hardware, the Wall Street Journal said it offers developers greater flexibility but could also be misused by hackers to identify software vulnerabilities without oversight.China pushes aheadThe Wall Street Journal also noted that Chinese cybersecurity company 360 Security Technology recently launched a fuzzing tool called Tulongfeng, which it said has bug-finding capabilities similar to Mythos."This kind of powerful weapon that can alter the landscape of cyberwarfare can't remain solely in American hands," 360 Security Chief Executive Zhou Hongyi told the Wall Street Journal.Anthropic has not publicly released Mythos over concerns about misuse. Under Project Glasswing, the company said the model found a 27-year-old vulnerability in OpenBSD and helped Mozilla identify 271 security vulnerabilities in an early version of Firefox.The Wall Street Journal said the narrowing gap between Chinese and US AI models is prompting companies to consider lower-cost alternatives, with firms including Microsoft exploring the possibility of offering Chinese AI models on their platforms.