- Chinese workers, students, and AI aficionados are regularly using leading American AI systems despite American companies’ attempts to limit their usage and increasingly fiery rhetoric from policymakers about the perceived risks of China’s access to American products.
- Chinese AI companies are challenging US labs and beating them on cost, with models like Moonshot's Kimi K3, which debuted on July 16, promising performance close to Anthropic’s Fable 5 at a fraction of the cost.
- DeepSeek, a Hangzhou-based lab, debuted its V3 and R1 models that matched the performance of US counterparts while claiming to have trained them on a tiny budget through engineering efficiency.
- Z.ai announced its GLM-5.2 model shortly after U.S. officials restricted access to Anthropic’s models for some users outside the U.S.
- Chinese AI models are becoming popular globally, with costs significantly lower than their US counterparts; for instance, one million tokens from DeepSeek-V4-Pro cost about $0.87 compared to $50 for Anthropic’s Fable model.
- AI users globally are adapting to the new reality where Chinese AI models are competitive with U.S. models on capability and far superior on price.
Chinese AI companies are rapidly advancing, with Moonshot, Z.ai, and DeepSeek emerging as formidable challengers to U.S. AI labs. Moonshot's Kimi K3, launched on July 16, is touted as the largest open-source model, performing comparably to Anthropic's Fable 5 at a fraction of the cost.2345
Independent benchmarks have placed K3 among the top three AI models, with one source even ranking it as the best available. Meanwhile, DeepSeek has introduced its V3 and R1 models, matching U.S. performance while operating on a significantly lower budget through innovative programming techniques.
The competitive landscape is shifting; at one point in mid-July, six of the top ten models were from Chinese firms, including Alibaba and Baidu. The cost disparity is stark: one million tokens from Anthropic's Fable costs $50, while DeepSeek's V4-Pro costs about $0.87, and Z.ai's GLM-5.2 costs $4.40. Even Kimi K3 is relatively affordable at $15.
This affordability has led U.S. startups and Fortune 500 companies to integrate these models into their operations, as seen with Coinbase halving its AI spending by utilizing Kimi and Z.ai's models. The trend underscores a significant shift in the AI landscape, where Chinese firms are not only competing but also leading in cost efficiency and innovation.
“DeepSeek's V4-Pro costs $0.87 per million output tokens versus $50 for Anthropic's Fable, and Chinese models claimed 6 of the top 10 spots in mid-July. Coinbase CEO Brian Armstrong said switching to Kimi and Z.ai's GLM halved the company's AI spending.”
