arXiv:2604.07190v2 Announce Type: replace-cross Abstract: We present a comprehensive adoption snapshot of the leading open language models and who is building them, focusing on the ~1.5K mainline open models from the likes of Alibaba's Qwen, DeepSeek, Meta's Llama, that are the foundation of an ecosystem crucial to researchers, entrepreneurs, and policy advisors. We document a clear trend where Chinese models overtook their counterparts built in the U.S. in the summer of 2025 and subsequently widened the gap over their western counterparts. We study a mix of Hugging Face downloads and model de
This report provides concrete data and a timeline (summer 2025) for a significant shift in the open language model ecosystem's leadership from Western to Chinese entities, validating observations of increasing Chinese AI capabilities.
A strategic reader should care as this indicates a material change in AI foundational model development, potentially impacting future technological dependencies, competitive landscapes, and national security strategies.
The dominance in crucial open-source AI development has definitively shifted from the US to China, challenging previous assumptions about innovation leadership and setting the stage for increased competition.
- · Chinese AI companies
- · Developers leveraging diverse open models
- · Countries seeking alternatives to US tech stacks
- · US AI companies (specifically in open models)
- · Western strategic planners assuming US leadership
- · Developers solely reliant on Western open models
Increased adoption and integration of Chinese open language models across global research and commercial applications.
Accelerated investment and focus on domestic open-source AI development in other nations to counter perceived Chinese dominance or US dependency.
The emergence of 'AI blocs' where geopolitical alignments dictate the preferred open-source model ecosystems and supply chains.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.LG