China's Zhipu is closing in on top U.S. AI models with Anthropic and OpenAI held back

Last week, Zhipu launched its GLM 5.2 artificial intelligence model, generating significant excitement akin to that surrounding DeepSeek's debut last year in Silicon Valley. This new open-source model has quickly gained traction, closely approaching Anthropic's Opus 4.8 on a vital agentic benchmark, all while being priced at about one-fifth of the competition. The response from developers has been robust, with traffic on OpenRouter surging even faster than after DeepSeek's V4 release in April. Unlike DeepSeek, which was ultimately regarded as a transient chatbot novelty, GLM 5.2 demonstrates impressive capabilities in crucial tasks such as planning, coding, testing, and looping—areas where businesses are eagerly looking to implement automation. With many companies grappling with unexpectedly high expenses related to AI tokens, which measure data processed and produced by AI systems, the focus has shifted to maximizing value for investment. The emerging preference for 'intelligence per dollar' makes Zhipu's economical yet effective model an appealing option. Gabe Pereyra, co-founder of Harvey, expressed his surprise at the rapid advancement of open-source AI, stating, "GLM 5.2 represents the first instance where it genuinely competes with established closed-source frontier models." Beneath the competitive pricing lies a transformative moment for open-source AI. GLM 5.2 is freely available for download, customization, and deployment on company servers, introducing competitive pressure on leading labs amidst increasing regulatory uncertainty. Recently, Anthropic had to withdraw its Fable Mythos-class model following a directive from the Trump administration, while OpenAI announced restrictions on its GPT 5.6 models due to a government request. This regulatory environment is making open-source solutions increasingly attractive as a reliable alternative.