A reporter based in London at The Verge, specializing in artificial intelligence coverage and serving as a Senior Tarbell Fellow, provides insights into the intensifying competition between Chinese and American tech firms in the AI landscape. Recently, Chinese companies Moonshot and Alibaba introduced advanced AI models that they assert can rival offerings from OpenAI and Anthropic at significantly lower prices. This rapid advancement signals a tightening of the competitive landscape, particularly as AI technology becomes increasingly pivotal in areas such as national security and economic influence.
Moonshot AI, a prominent AI developer in Beijing, kicked off this wave of announcements by launching its model, Kimi K3, last Friday. The company claims that its internal evaluations consistently position Kimi K3 above most US systems, only trailing behind OpenAI’s GPT-5.6 Sol and Anthropic’s Claude Fable 5, with Kimi K3 performing better on certain benchmarks.
Shortly after, Alibaba previewed its own model, Qwen3.8, which it describes as "one of the most powerful model[s] available today" and claims it comes in just behind Fable 5, Anthropic’s leading model.
A significant distinguishing factor for both Chinese firms is their commitment to public accessibility. Unlike some US companies, including Meta, which have begun releasing models for developers to download and modify, Moonshot and Alibaba are offering their advanced models openly, enhancing their competitive stance within the AI sector.
Moonshot refers to Kimi K3 as the largest open-source AI system globally, boasting an impressive 2.8 trillion parameters. This metric serves as a rough gauge of a model's complexity and performance potential, although a higher parameter count does not always equate to superior capabilities. Meanwhile, Alibaba’s Qwen3.8 features a model size of approximately 2.4 trillion parameters and is described as "continuously evolving." In contrast, OpenAI and Anthropic do not publish specific parameter counts for their top models.
The true performance of these Chinese models remains uncertain until they are fully launched and assessed independently. Moonshot has announced plans to release the complete model weights on July 27, while Alibaba indicates that Qwen3.8 will also be offered with open weights shortly.
The introduction of these models heightens the competitive dynamic between the United States and China in what many consider one of the most significant technological contests of the era. The impact of these developments is reminiscent of last year’s launch of a cost-effective model by DeepSeek that posed a serious challenge to US systems. Moreover, the emergence of these models raises critical questions about whether the substantial investments made by American firms in infrastructure and model development can guarantee a lasting edge, especially when Chinese competitors may achieve similar or greater capabilities with fewer resources.
The release of these advanced Chinese models, which can be freely downloaded and customized, starkly contrasts with the more cautious approach of American labs that maintain proprietary systems. This trend of openness persists despite the U.S. government’s rapid moves to limit global access to top-tier technology, such as implementing export controls to block China's access to leading chips and compelling Anthropic to withdraw its advanced systems from international markets due to concerns regarding competition.
While the true efficacy of these new Chinese models is yet to be verified, their emergence is poised to escalate the technological rivalry between the United States and China, influencing economic strategies and security policies, and revealing that America’s lead in AI may not be as substantial as previously thought.
