China has taken a significant leap in its pursuit of AI-driven scientific advancement with the launch of “ScienceOne,” a groundbreaking artificial intelligence model designed to dramatically accelerate research across diverse fields. Unveiled by the Chinese Academy of Sciences (CAS) at the 2025 World Artificial Intelligence Conference, ScienceOne promises to reshape fundamental research methodologies and expedite the pace of discovery.
Developed jointly by twelve CAS institutes, ScienceOne is engineered to comprehend complex scientific data modalities, including waveforms, spectra, and fields. Its core capabilities integrate literature extraction, sophisticated knowledge reasoning, and the orchestration of computational tools, aiming to provide an intelligent foundation for technological innovation.
Current approaches to AI in science often involve fine-tuning general-purpose models or developing isolated, single-field tools, which face limitations due to fragmented data systems, insufficient specialized reasoning, and closed research ecosystems. ScienceOne directly addresses these challenges by building upon China’s open-source foundation models, deeply customized for scientific applications. It incorporates proprietary models for common scientific data types and integrates specialized tools like AlphaFold for protein folding and MatterGen for materials generation.
Early evaluations have demonstrated ScienceOne’s state-of-the-art performance in disciplines spanning mathematics, physics, chemistry, materials science, and biology, showcasing leading capabilities in scientific tool invocation and reasoning. Remarkably, the model also achieved excellent results in Humanity’s Last Exam (HLE), a rigorous benchmark for scientific understanding.
Beyond its foundational capabilities, the research team behind ScienceOne has developed two intelligent agents. One agent is designed to assist researchers with comprehensive literature reviews, article writing, and evaluating research topics and technical pathways. With access to 170 million pieces of scientific literature and real-time open-source information, this agent can reduce literature review tasks that previously took days to mere minutes. The second agent streamlines the use of scientific tools, autonomously planning and scheduling over 300 computational tools, significantly enhancing research efficiency.
ScienceOne is already making an impact, powering the X-Cell digital cell platform for automated biological target identification, improving particle simulation efficiency at the Beijing Electron-Positron Collider, enhancing molecular prediction accuracy in chemistry, and streamlining operations for global telescopes. This launch marks a pivotal moment in China’s strategy to leverage AI as a catalyst for scientific breakthroughs, promising to accelerate discoveries that address humanity’s most pressing scientific challenges.
