The current balance of power in open models (www.interconnects.ai)

A recent briefing highlighted the shifting landscape of open-weight AI models, particularly in the context of U.S.-China competition. Open-weight models, which allow public access to their underlying weights for inspection and use, have seen significant advances, especially from Chinese companies. As of now, Chinese models like Alibaba's Qwen and Z.ai's GLM series have not only surpassed their American counterparts in download metrics but are also leading on performance benchmarks, raising concerns about the U.S.'s competitive position in AI development. The statistics reveal that China's open-weight models have amassed over 3.2 billion downloads since early 2025, reflecting a growing dominance in accessibility and capabilities within the AI community. This competitive edge stems from rapid deployment cycles and focused task distributions that cater to user demand, particularly in areas like agentic coding. While American firms are beginning to counter this trend with releases and partnerships, the gap still remains significant, with American models often lagging by six to nine months in performance compared to closed models from companies like OpenAI. As the utilization of these models expands—evident from their increasing adoption in academic research and enterprise applications—there's urgency for the U.S. to bolster investment in open-weight models to retain relevance and foster a safer AI ecosystem amid rising global competition.