New open-weight artificial intelligence models from American and European startups are struggling to match the performance of their Chinese counterparts, despite pouring more resources into development, according to experts who spoke at a Rest of World event in San Francisco.
Practical Challenges in Developing Nations
For businesses and governments in lower-income countries, cost and the expertise required to tailor models for specific applications in sectors like education, healthcare, and agriculture are key considerations.
Daniel Björkegren, an assistant professor at Columbia University, noted that American models are heavily optimized for American workloads. “When you take those same models and try to deploy them in remote parts of the world, often they don’t work so well,” he said. This limitation has prompted some users to switch models when new capabilities or lower prices emerge, though Björkegren raises a critical question: the ease of switching remains a key concern.
Read Also: Meta uses influencers to promote teen safety tools globally
Chinese Models Surge Past US Rivals
Since the release of DeepSeek early last year, Chinese open models have rapidly closed the gap with U.S. rivals. A review by OpenRouter found four Chinese models among its top five last week, including DeepSeek, Z.ai, Xiaomi, and Tencent. LiveBench, a model ranking platform, recently placed DeepSeek in the top five for the first time, showing Chinese models lag behind U.S. counterparts by just 3% on benchmark scores. Graham Webster, a research scholar at Stanford’s DigiChina Project, stated, “If you need relatively frontier capability, right now that’s Chinese open models only.”
China’s leadership views open models as central to its goal of global AI dominance by 2030. The country has established the World Artificial Intelligence Cooperation Organization to promote AI adoption and shape governance in developing nations. Kristy Loke, a research fellow at the AI safety program MATS, explained that open models help China sidestep trust issues tied to companies like Huawei, which faced U.S. restrictions. “With open models, because people can download them and use them and own them, it’s private to them — so you kind of break down the trust problem,” she said.
US Restrictions Push AI Diversification
The U.S. government’s directive to Anthropic to restrict foreign access to its models like Fable 5 and Mythos on national security grounds has also driven governments to diversify their AI dependencies. Björkegren emphasized that relying on a single foreign provider creates risks if access is revoked for geopolitical reasons. “Strategically, a government is going to be in a better position if there are more providers from different countries,” he argued.
Read Also: BYD Expands Fleet to Cut Shipping Costs Amid Global Challenges
Recent incidents involving AI agents from OpenAI, Anthropic, and Meta hacking websites have shifted perceptions of open models as inherently safer. Hugging Face reported using a Chinese open-weight model to defend against AI agents from OpenAI, which refused its prompts. Trade restrictions limiting China’s access to Nvidia’s most advanced chips have pushed Chinese firms to develop more efficient models, a feat Western companies have yet to match.
The Chinese government aims to sustain this lead, with Webster noting that domestic and global open models are critical for soft power and economic growth. “They can throw more money at things, but you can only buy so many chips under current conditions,” he said. “The Chinese companies have been doing a very good job of figuring out how to do more with less.” The efficiency of Chinese models stems from restrictions on high-end Nvidia chips. Trade limits forced companies to optimize software and algorithms, achieving comparable performance with fewer resources. This efficiency translates to significant cost savings, with some tasks running 50% to 90% cheaper than American systems.