China Strikes Back Against U.S. AI Supremacy

China’s top AI firms are intensifying the competition with Silicon Valley, as Moonshot and Alibaba have introduced models they assert can rival the best offerings from OpenAI and Anthropic at a significantly lower cost. These swift developments indicate that the U.S. is facing increasing pressure in the AI arena, especially as this technology becomes crucial for national security, economic strength, and global influence.

The first move came from Moonshot AI, a prominent developer based in Beijing, which launched Kimi K3 recently. According to Moonshot, their testing shows Kimi K3 consistently outperforms nearly all U.S. systems, only falling short of OpenAI’s GPT-5.6 Sol and Anthropic’s Claude Fable 5, although it excelled in certain benchmarks.

Shortly after, Alibaba showcased Qwen3.8, a new model it claims is “one of the most powerful models available today” and “second only to Fable 5.”

Both companies are highlighting a significant distinction from leading U.S. labs: instead of keeping their most advanced models under wraps, they are making them publicly accessible. While some U.S. companies, like Meta, have adopted a similar strategy by releasing models for developers to download and modify, this trend of openness is becoming a defining feature of China’s AI sector.

Moonshot touts Kimi K3 as the largest open-source AI system globally, boasting 2.8 trillion parameters. Parameter counts serve as a measure of a model’s complexity during training and provide a rough gauge of its scale and performance, though a higher count doesn’t always equate to better results. Alibaba claims Qwen3.8 has 2.4 trillion parameters and is “continuously evolving.” In contrast, neither OpenAI nor Anthropic discloses the exact parameter counts for their leading models.

Evaluating the true capabilities of these Chinese models will be challenging until they are fully released and independently assessed. Moonshot plans to release the complete model weights—numerical data learned during training—on July 27th, while Alibaba states that Qwen3.8 will be “going open-weight soon.”

Despite this, the announcements have already intensified the rivalry between the U.S. and China in what is often described as the defining tech race of our era. This shift has disrupted the industry in a manner reminiscent of last year when DeepSeek introduced a low-cost model that competed with leading U.S. systems. The emergence of these models raises questions about whether the substantial investments U.S. companies are making in chips, data centers, and model training can maintain a lasting edge, especially if Chinese competitors can reach or exceed that level with fewer resources.

The potential release of two highly capable Chinese models for public use starkly contrasts with the more secretive approach of U.S. labs, where the most advanced systems remain proprietary. This openness comes at a time when the U.S. government is rapidly implementing measures to limit global access to the underlying technology. Export controls have been used to restrict China’s access to cutting-edge chips and to compel Anthropic to withdraw its most advanced system from the market due to concerns about aiding foreign competitors.

Whether these new Chinese models will meet their creators’ claims is yet to be determined. However, similar to DeepSeek, they are poised to heighten the technological competition between the U.S. and China, impacting economic and national security policies, and revealing that America’s lead is narrower than it once seemed.