The Beijing Spectrometer — a state-of-the-art particle detector often described as a "high-speed camera" for microscopic particles invisible to the naked eye — is entering a new phase of intelligent, autonomous and more efficient operation, reflecting what researchers call agentic artificial intelligence.

Over its 16 years of operation, the large-scale scientific facility has generated vast amounts of data. Combined with the complexity of its research targets, that has led to long analysis cycles and challenges for traditional, human-led analysis methods, said Liu Beijiang, a researcher at the Institute of High Energy Physics of the Chinese Academy of Sciences.

Now, guided by what researchers describe as an "intelligent brain," automated scientific workflows can integrate AI-enhanced tools to reduce repetitive tasks and improve efficiency, Liu said.

The "intelligent brain" is part of a suite of eight large language models tailored to different scientific disciplines, along with a foundational scientific model, released by the Chinese Academy of Sciences in Beijing on Tuesday. The system, called ScienceOne 100, is designed to advance "AI for Science" — shifting research from fragmented, isolated efforts toward a more integrated, platform-based model of innovation.

Large language models are AI systems trained on vast amounts of text and data that can understand, generate and analyze language. In this case, they are adapted for scientific research tasks.

"With the acceleration of a new round of technological revolution, AI for Science has shifted the paradigm in scientific research. It is redefining the innovation landscape of the new era, as well as enhancing the efficiency and quality of scientific development," said Zeng Dajun, a researcher at the CAS' Institute of Automation.

China's 15th Five-Year Plan (2026-30) highlights the importance of fully implementing "AI Plus," an initiative that uses artificial intelligence to process complex scientific data, improve research and development efficiency, and support more collaborative human-machine research methods.

Zeng said the system, built on a foundational scientific model first launched in July 2025 and continuously improved since then, promotes open sharing of data, models and algorithms to support scientific innovation.

He said its most widely used function is an intelligent literature assistant, which helps researchers analyze academic papers. According to CAS, it improves accuracy from about 70 percent in mainstream international systems to 90 percent through cross-checking and multi-step validation, helping reduce errors known as AI hallucinations, where systems generate incorrect or misleading information.

The system also allows users to intervene during operation, adjust tasks in real time, reduce trial-and-error work and shorten research time by more than 60 percent, Zeng said.

While the foundational model addresses broad scientific problems and supports interdisciplinary research, the eight discipline-specific models focus on more specialized challenges, he said. Together, they are designed to create a coordinated research ecosystem across different fields.

Liu Jianjun, a researcher at the CAS' Shanghai Institute of Ceramics, highlighted the model's role in developing high-performance materials used in extreme environments.

Such materials are essential for advances in aerospace and integrated circuits, but traditional development cycles can take about 15 years due to the large number of variables involved, he said.

By integrating scientific literature, physical laws and material properties, the model can help generate customized material designs. Liu said it also improves the link between laboratory research and real-world application, speeding up development.

Yang Yanchu, a researcher at the CAS' Aerospace Information Research Institute, said a large language model for the near-space domain fills a gap in China's research capabilities.

Near space, at altitudes of 20 to 100 kilometers, is a transitional region between the atmosphere used by aircraft and the orbital environment of satellites.

Yang said the model simulates atmospheric, meteorological and electromagnetic conditions in that region and supports applications such as communications, navigation, remote sensing and detection. It also integrates knowledge in energy, materials and flight control.

Compared with other models of similar size, Yang said it performs better in logical consistency, accuracy and technical reasoning, providing more reliable expert-level responses.