SHANGHAI, July 17 (Xinhua) -- An upgraded 2.0 version of the ScienceOne Omni scientific foundation model was launched on Friday at the 2026 World Artificial Intelligence Conference and High-Level Meeting on Global AI Governance in east China's Shanghai, according to the Chinese Academy of Sciences (CAS).
Developed by a joint research and development (R&D) team led by the CAS, the ScienceOne Omni serves as a dedicated intelligent foundation for scientific tasks. It aims to break down barriers between general artificial intelligence (AI) and specialized scientific capabilities across various domains, providing more reliable and comprehensive intelligent support for scientific study, said the CAS.
The ScienceOne Omni was launched in July 2025. The newly launched 2.0 version introduces multiple upgrades, enabling this scientific foundation model to perform understanding, reasoning, prediction and generation across different scientific modalities and tasks.
In terms of data support, its R&D team constructed 8 million high-quality scientific reasoning data entries covering more than 200 tasks, training the model to conduct reasoning based on chains of knowledge and evidence.
Regarding computing infrastructure, the model has established a heterogeneous computing system integrating supercomputing, intelligent computing and rapid computing. Leveraging advantages of multiple scientific computers, it can swiftly complete high-precision simulation tasks for fields such as molecular dynamics and quantum chemistry, generating high-quality scientific data to support model training, evaluation iterations and study applications.
Based on this foundation model, the R&D team has built an integrated scientific research intelligent platform to provide common intelligent services for application scenarios across a variety of fields and support researchers in customizing specialized models and scientific intelligent agents on demand. Multiple intelligent agents developed based on the model have also been deployed at scale and are undergoing upgrades, the CAS noted.
To date, the ScienceOne Omni scientific foundation model has been applied to multiple research fields, including mechanics, astronomy and chemistry. Moving forward, the research team will continue to expand application scenarios and explore more possibilities for AI to empower scientific research, said the CAS. ■
