Several reasons explain the price difference. The first is the pricing mechanism. For closed US models, the model owner controls the API, deployment infrastructure, security layer, compliance system, storage capabilities, tools, connectors, availability guarantees, and enterprise product. The price is largely determined by the company. Open weight models are priced more competitively and prices change more frequently. Many providers can host the same model, optimize inference, and compete on price. A model that can be hosted by many providers will face greater pricing pressure than a model that can only be accessed through one owner.
Second, Chinese AI models have lower training costs because they have no margin for error like US companies do. Export controls and weaker access are effective because they discourage Chinese labs from building their own computing infrastructure. That’s why Chinese labs are working with distillation of US models, open-weight practices and inference efficiency, reducing the number of expensive experiments required before publication. This suggests that model safety testing is more limited compared to US models.
Third, the strategies on both sides are different. Many Chinese labs use open weight releases to build adoption, prestige, and developer ecosystems. A cheaper model can spread more quickly. It can become the standard for developers who don’t need the absolute best boundary model. It can also build global influence even if the company doesn’t control every stake. In contrast, US border AI is now led by closed-source models, as the United States is a leader in computation and therefore does not rely on open-source models.
Q3: Do open weight models provide a better strategy?
A3: Not necessarily. There are many advantages and disadvantages. The main advantage is the speed of diffusion. Open-weight models can be distributed via Hugging Face, GitHub, cloud providers, on-premises deployments, and third-party inference platforms. A European startup, a Southeast Asian government agency, or a Latin American developer can use Qwen, DeepSeek, GLM, or Kimi through a third-party host without sending logs directly to the original Chinese lab. This reduces political and operational hurdles to implementation. Hugging Face recently announced that Chinese models surpassed U.S. models in both monthly and total downloads on their platforms, and that Chinese models accounted for 41 percent of their downloads compared to last year. This helps build China’s reputation and weakens the perception that the United States has an unassailable lead. There is also a de facto technical dependency on Chinese model families, even if the models are hosted outside of China. Many users don’t care whether the model is Chinese or American as long as they can host it locally.
Open-weight models are also partly a necessity for China. US export controls limit access to the best AI chips, and Chinese labs don’t have the same global inference capacity as US hyperscalers. If a Chinese lab had to serve a global user base entirely with its own computing power, it would face a major capacity problem. Open weight releases reduce this burden because third parties and local users provide much of the computing power provided. The model can be distributed worldwide without every request having to come through China.
But open weight models cut the other way. Chinese labs do not collect global user data in the same way when their models are used through third-party hosts. When a model is downloaded from a US cloud provider, run through a third-party inference platform, or deployed locally to an enterprise, the original Chinese developer cannot see the prompts, logs, feedback, tool calls, or product behavior. In other words, monetization is extremely difficult with the open weight strategy.
In the larger context, these different strategies will shape the future of AI competition. Models improve when companies can see how users interact with them, where they fail, what tools they connect to, what tasks users repeat, and what outcomes users prefer. US companies have an advantage here because they build integrated AI products around ChatGPT, Claude, Gemini, Microsoft Copilot, enterprise APIs, coding agents and cloud platforms. Integrated products create feedback loops, corporate relationships, and product loyalty that cannot be achieved through the distribution of open weights alone. As Chinese open-weight models spread globally through third-party hosting and local delivery, Chinese labs like Alibaba, Tencent, Baidu and ByteDance will face the challenge of monetizing the same global engagement data, subscription revenue and platform ties that closed US providers can build.
Q4: What are the implications for US politics?
A4: The United States faces a two-part problem. It needs to both protect its border advantage and convince the rest of the world to build on America’s AI stack. The Trump administration’s strategy to export the AI stack around the world is moving in the right direction. The American AI Exports Program aims to promote full-stack AI packages abroad. The Commerce Department’s request for proposals broadly defines these packages, including AI-optimized hardware, data pipelines and labeling systems, AI models and systems, cybersecurity measures, and industry-specific applications. Such measures are also a strategic response to the spread of open weights in China. If global players can build on U.S. chips, clouds, models, cybersecurity standards and applications, the United States will also lead the way across the ecosystem.
But this strategy relies on trust. The United States cannot require other countries to build out their AI stack while giving them the impression that access can be suddenly and unilaterally changed. The recent model access ban was not a good sign. On June 12, 2026, the US government blocked Fable and Mythos from foreign access, resulting in Anthropic’s complete withdrawal of the models. If foreign companies believe they can quickly lose access to the U.S. model, they will diversify. Some will opt for Chinese open weight models. Some will choose local sovereignty models. Some will use multiple providers to avoid dependence on Washington. All this computing power without a global user base will not give the United States the advantage it expects.
Prices are also a challenge. Leading US models remain expensive for many developers and governments, while Chinese open-weight models represent a cheaper alternative for many companies. As the gap between U.S. closed-source models and Chinese open-weight models narrows, this price difference will likely lead to a push toward open-weight models in the United States as well. AI companies therefore have to adjust their pricing structure. As far as policymakers are concerned, the U.S. export strategy should focus on reducing prices and making U.S. models available through lower-cost options.
Overall, recent Chinese model releases show that the gap in AI model capabilities is small. For this reason, the U.S. AI strategy should both gain global acceptance and lead through frontier models. The United States still has major advantages, but Chinese models are now powerful enough, cheap enough and open enough to shape global AI competition. For this reason, winning the global AI competition will depend on the ability to build trust as a reliable provider.
Yasir Atalan is deputy director and data fellow of the Futures Lab at the Center for Strategic and International Studies (CSIS) in Washington, DC
https://www.csis.org/analysis/what-know-about-chinese-ai-models
