Jefferies' assessment of the GLM-5.2: It ranks among the world's top tier, but is still far from replacing Anthropic.

According to Mars Finance, on June 22, Jefferies stated that Zhipu's GLM-5.2 entering the top three global large model rankings is a significant milestone in China's AI development. However, the US restrictions on access to Anthropic's top models are unlikely to bring significant revenue growth to Zhipu. In a China technology industry report released on June 17, Jefferies stated that GLM-5.2 ranked third in the global Model Intelligence Index, behind only Anthropic and OpenAI. This marks the first time a Chinese model has entered the global top three. The report stated that GLM-5.2 excels in programming and long-term agent workflows, becoming one of China's top-ranked large models. Zhipu released GLM-5.2 through its Coding Plan on June 13 and opened its API on June 17. The model uses a 744 billion parameter MoE architecture, with approximately 40 billion active parameters, supports 1 million token context windows, and has its weights open under the MIT license. Jefferies reports that GLM-5.2 ranks third globally and first in China in Artificial Analysis's Model Intelligence Index; it ranks fourth globally and second in coding and first in agentic metrics, respectively. In the front-end programming rankings, GLM-5.2 Max ranks second, only behind Fable 5 High, and higher than Claude Opus 4.7 and 4.8. The report also points out that the US restrictions on access to Anthropic's Claude Mythos 5 and Fable 5 may be unsustainable in the long term. After Anthropic released its models on June 9th, the US government, citing national security concerns, required it to restrict access for foreigners, including non-US employees within the US. Due to the difficulty in distinguishing user identities in the short term, Anthropic temporarily suspended global access and strengthened identity verification. Jefferies believes that this arrangement of "limiting access to cutting-edge models to Americans only" is difficult to implement and may harm the US AI ecosystem. Many AI researchers and engineers in the US are foreign-born or non-US citizens. Excluding them from cutting-edge models could slow research progress and even drive some talent and usage demand overseas. However, the report argues that Fable 5's limitations will not significantly translate into revenue for Chinese model companies. Jefferies' channel research shows that most users will not directly switch from Claude to GLM-5.2. GLM-5.2's programming capabilities are considered close to Claude Opus 4.7, but still lag behind Opus 4.8. Even if developers adopt Chinese open-source models, they may use them through local deployment, cloud vendors, or inference platforms like OpenRouter, rather than necessarily contributing revenue to model developers. The report also states that developers typically mix and match models for different tasks, including code generation, debugging, and long-context tasks, rather than completely migrating to a single model. This limits GLM-5.2's commercialization potential for Anthropic traffic spillover. Jefferies points out that Zhipu still faces insufficient high-end inference computing power, especially since long-context and agent workflows consume more computing power, potentially affecting the delivery of enterprise-level requirements. The report also mentions that the Chinese large-scale model market is highly competitive in terms of price. While GLM-5.2 offers approximately 26% better performance than GLM-5.1, its API price remains unchanged, indicating a lack of significant pricing power for model manufacturers. Furthermore, the dedistillation technology and export controls imposed by US model manufacturers pose long-term risks. Jefferies states that these measures could limit the ability of Chinese models to catch up with the forefront and lead to a widening of the performance gap between Chinese and US models. The report's core conclusion is that GLM-5.2 proves that Chinese models have entered the global top tier, but it has not yet truly "put on Anthropic's shoes." Constrained by factors such as revenue conversion, computing power supply, price competition, and US technological restrictions, Zhipu is unlikely to gain substantial commercial growth from Anthropic access restrictions in the short term.