Media: China's AI model GLM-5.2 challenges U.S. giants at low cost, rapidly rising to top ranks on popular model lists
The newly released large language model GLM-5.2 from Chinese AI startup Zhipu AI has quickly drawn international attention with performance approaching that of leading American AI models and significantly lower usage costs, reigniting discussions about China’s accelerating advancement in artificial intelligence relative to the United States. However, analysts believe data security remains the primary obstacle for this model entering enterprise markets in Europe and the U.S.
According to Reuters, the global market was shaken earlier this year when DeepSeek launched a high-performance, low-cost large model. Now, Zhipu AI’s recent release of GLM-5.2 has been dubbed by some industry insiders as "another DeepSeek moment."
Data from developer platform OpenRouter shows that GLM-5.2 has rapidly climbed to the upper ranks of popular model lists, thanks to its strong capabilities in code generation and AI agent functions—its current ranking now surpasses certain models from U.S.-based AI company Anthropic.
Industry experts note that GLM-5.2 can complete complex programming and task execution at a cost far lower than comparable U.S. products, with performance nearing that of top-tier models from OpenAI and Anthropic.
Brian Tse, founder and CEO of Concordia AI, a Beijing-based AI safety advisory firm, said international developers are increasingly aware of the risks involved in relying solely on U.S.-developed AI models. He stated: "The international developer community is becoming more conscious of the clear risks associated with depending exclusively on large models developed in the United States."
Analysts point out that GLM-5.2’s positive reception in global markets also reflects growing corporate interest in open-source AI models. As AI application costs continue to rise, more companies are turning toward lower-cost open-source alternatives. In contrast, closed-source AI agent models typically consume more computing resources and tokens, thereby increasing operational expenses.
Recent rankings from the AI evaluation firm Artificial Analysis place GLM-5.2 fifth globally in overall performance among large language models; it ranks second in front-end programming ability on the Code Arena code benchmark platform, while operating costs are only about one-sixth those of mainstream U.S. closed-source models such as Anthropic’s Claude series and OpenAI’s GPT series.
Nevertheless, analysts caution that data security remains the main barrier to wider adoption of GLM-5.2 in European and American enterprise markets.
Wei Sun, Chief AI Analyst at market research firm Counterpoint Research, noted that some European enterprises have begun evaluating whether GLM-5.2 is suitable for enterprise applications. However, in both the EU and the U.S., certain clients, partners, and regulated industries may still reject integrating Chinese AI models into their infrastructure due to data security concerns—regardless of technical superiority or pricing advantages.
Some experts argue that concerns over data security related to Chinese AI models are exaggerated. If deployed on U.S. cloud platforms or on-premise enterprise servers, data security can be effectively ensured. While large enterprises are adopting these models at a relatively slow pace, tech startups and small-to-medium enterprises are embracing them much more quickly.
Source: rfi
Original article: toutiao.com/article/1869666032792586/
Disclaimer: This article represents the personal views of the author