Who Uses Open-Weight Models? China and the Shifting Geography of AI in Science

A research study is underway to analyze the adoption of open-weight large language models (LLMs) in scientific research. The study, which examines 21 million full-text articles, aims to determine if researchers are actively using open-weight models, moving beyond mere mentions. It also seeks to differentiate between their use in applied versus foundational AI research.

Why it matters

Understanding the adoption of open-weight models is critical for assessing the trajectory of AI development and its diffusion across research domains. This insight can inform resource allocation for model development, infrastructure, and policy frameworks governing AI research and collaboration.

Key insights

• The research investigates the adoption patterns of open-weight models in scientific research.
• The study analyzes 21 million full-text articles from the Semantic Scholar Open Research Corpus (S2ORC) up to June 2026.
• A mixed Natural Language Processing (NLP) pipeline is employed to identify model occurrences and distinguish between use and mention.
• The corpus is categorized into single- and multi-model family studies to proxy for applied and foundational AI research, respectively.

Source

arXiv — Computers and Society — https://arxiv.org/abs/2608.11090