trailed their leading U.S. rivals

New open-weight AI models from American and European firms are targeting Chinese counterparts to address security concerns, though they struggle to match performance despite better resources, according to experts speaking at a Rest of World event in San Francisco. Last week, American startup Reflection unveiled its inaugural open-weight model, with European labs Aleph Alpha and Mistral also releasing new open-weight models.

Open models provide users greater customization and cost savings compared to closed models from OpenAI, Anthropic, and Google. For businesses and governments in lower-income nations, cost and customization expertise are the main considerations. Assistant Professor Daniel Björkegren from Columbia University emphasized that in low-income contexts, users worry about dependency on the U.S. for advanced AI intelligence, as American models are often optimized for American workloads and may not perform well in remote areas.

Chinese open models have recently narrowed the performance gap with their U.S. rivals, with several U.S. startups, businesses, and the European Union’s sovereign AI project adopting them. The OpenRouter leaderboard shows that four Chinese models were among the top five last week. China's President Xi Jinping views open models as crucial for the country's goal of becoming a global AI leader by 2030, and China has established a World Artificial Intelligence Cooperation Organization to promote AI adoption and governance in developing nations.

Critics argue that China's open models could help the country circumvent concerns about its presence in overseas markets. The U.S. government's recent directive to Anthropic to suspend access to its Fable 5 and Mythos models for foreign nationals also highlights the need for nations to diversify their AI providers to avoid relying on foreign systems.

Chinese AI companies face compute limitations and monetization challenges, but their open models can be up to 90% cheaper than leading American models. Some Chinese firms, like Moonshot AI, are negotiating revenue-sharing agreements with U.S. cloud providers to access their models.