ScienceOne was officially released at the 2025 World Artificial Intelligence Conference (WAIC) on July 26. [IMAGE: WAIC]

An artificial intelligence platform — ScienceOne — of the Chinese Academy of Sciences (CAS) was officially released at the 2025 World Artificial Intelligence Conference (WAIC) on July 26. It has achieved an in-depth understanding of various scientific modal data such as waves, spectra, and fields, and possesses core capabilities including the extraction and integration of scientific literature, the representation and reasoning of scientific knowledge, and the arrangement and planning of scientific tools.

ScienceOne, developed by CAS’s Institute of Automation and partners, efficiently manages resources such as data and models, as well as schedule various tools such as computational simulation. Researchers can seamlessly integrate the model throughout their research process, achieving an imperceptible embedding of artificial intelligence in scientific research.

In terms of core architecture, built on domestic open-source large models, ScienceOne is tailored for the scientific field and integrates a series of self-developed dedicated models for common scientific data modalities, alongside professional models from AlphaFold and MatterGen.

In terms of scientific capabilities, this large model has systematically mastered the core theorems, laws and professional knowledge of the six major disciplines of mathematics, physics, chemistry, astronomy, geology and biology, and has achieved in-depth understanding of various scientific modal data such as waves, spectra and fields.

Based on ScienceOne, the team has also developed two scientific agents, namely S1-Literature and S1-ToolChain.

S1-Literature of ScienceOne has integrated 170 million scientific and technological literature and real-time open-source scientific and technological information, enabling in-depth understanding of scientific data containing formulas and charts. With its support, the literature research work that previously took 3 to 5 days to complete can now be completed in just 20 minutes.

S1-ToolChain of ScienceOne can independently plan and invoke over 300 scientific computing tools, achieving collaborative arrangement and convenient invocation of tools. It can automatically identify research tasks, intelligently arrange and schedule the optimal tool chain, enhancing the efficiency of the research process. It supports users in flexibly integrating their own agents and tools to quickly build dedicated research applications.

At present, ScienceOne has been applied in multiple disciplinary fields such as life sciences, high-energy physics, and mechanics research.

In addition, ScienceOne is helping scientists enhance experimental efficiency in chemical synthesis, achieve more accurate prediction results in molecular structure prediction, and realize intelligent global resource scheduling and analysis in astronomical observations.

In the process of serving real scientific needs, ScienceOne will continue to be iterated to enhance its practicality and reliability.

Source: Institute of Automation,

Chinese Academy of Sciences