There is an analysis that China's artificial intelligence (AI) technology level is evaluated to have caught up to the level of Antropic or Open AI.
In particular, 'GLM-5.2', recently released by China's 'Z.ai' (formerly Zipoo AI), has performance comparable to Antropic's cutting-edge model, which is considered the best so far, but the usage fee is much cheaper at one-sixth of the level, causing a stir in the industry again.
The GLM-5.2 model was released just a few days after the U.S. government ordered Antropic's latest models 'Fable 5' and 'Mythos 5' to be banned from use by foreigners two weeks ago, citing security issues.
Rehan Ahmad, co-founder of Silicon Valley startup Alphasive, said after using the GLM-5.2 model for more than a week, “With Fable 5 being regulated, the technology gap between the U.S. and China has narrowed significantly.”
In fact, the GLM-5.2 model has the advantage of being much cheaper and can be used without restrictions in the United States, entering the top 10 of the world's most popular AI models.
Six Chinese models are included in the ‘Top 20’ models of Artificial Analysis, a global AI analysis company.
The New York Times (NYT) reported on the 25th (local time) that the low price is a factor that overcomes obstacles to China's AI model being widely used in the United States, such as concerns about its connection with the Chinese government and criticism of copying the model.
18 months ago, Chinese startup Deepseek surprised Silicon Valley by introducing cost-effective AI. Z.ai is also pursuing a strategy of providing services at a much lower price than the US model.
OpenRouter, a startup that operates an AI model leaderboard, said that when performing certain tasks, GLM-5.2 operates at about one-eighth the cost of Entropic's 'Opus 4.8'.
Like other Chinese models, GLM-5.2 is open source software, so anyone can use and modify it for free. This means that although it may not be as powerful as the American product, it can be used much cheaper.
“Do people have to drive luxury cars wherever they go? Probably not,” said Vivek Ramaswamy, an early investor at venture capital firm Madrona Venture Group.
The GLM-5.2 model is particularly good at generating computer code and running AI agents that use other software to perform tasks.
The American industry is looking at the rapid growth of Chinese models with suspicion and caution.
Antropic claims that Chinese AI research institutes, such as Alibaba's Q1, mobilized tens of thousands of fake accounts to collect data and imitate technology (distillation) by asking tens of millions of questions to its 'Claude' model.
The U.S. Department of Commerce has already placed Z.ai on its blacklist (export control list) due to national security concerns.
So-called ‘distillation’, which uses data from one system to train another system, is often used in AI development. However, Antropic and OpenAI prohibit secretly collecting data for distillation. It is unclear whether Z.ai used distillation techniques in developing the new model.
Some say that it is impossible to build a top-level AI model through ‘distillation’ alone.
“It’s not true that all the features of the Chinese model come from Antropic,” said Charles O’Neill, head of model training at Basseten, a company that sells access to GLM-5.2.
Professor Jeffrey Ding of George Washington University said, "There were predictions that the U.S. government's export controls would eventually widen the gap between the latest American models and Chinese models, but GLM-5.2 is actually leading the situation in the opposite direction."
