Panshi: The Scientific Foundation Large Model — AI Empowerment Reshaping Paradigms and Unlocking Infinite Scientific Possibilities}
The Chinese Academy of Sciences unveils Panshi, a groundbreaking scientific large model that integrates AI with science, accelerating discovery and transforming research across multiple disciplines.
Artificial intelligence is reshaping the fundamental logic of scientific research and accelerating discovery. “AI + Science” has become a trend, offering unprecedented opportunities to solve major scientific and technological challenges faced by humanity.
On July 26, at the 2025 World Artificial Intelligence Conference, a Chinese team from the Chinese Academy of Sciences officially released the “Panshi·Scientific Foundation Large Model.” This intelligent base is trained with specialized scientific knowledge and data, capable of deeply understanding multiple scientific modalities such as waves, spectra, and fields. It features core capabilities like scientific literature extraction and fusion, knowledge representation and reasoning, and scientific tool orchestration, aiming to provide “rock-solid” intelligent support for technological innovation across fields.
Leveraging systemic advantages to build an “AI + Science” operating system
Currently, “AI + Science” research generally involves fine-tuning general large models with domain data and building dedicated tools for specific fields, facing challenges like data silos, limited reasoning capabilities, and closed research ecosystems.
Focusing on these issues, the joint team from the Chinese Academy of Sciences leverages a complete natural science discipline system, a full-stack AI innovation chain, major scientific facilities, and data resources to promote platform-based, systemic transformation of “AI + Science.”
The “Panshi·Scientific Foundation Large Model” manages resources like data and models, and orchestrates scientific tools such as simulations, enabling the entire research process from hypothesis generation, plan formulation, simulation, validation, to pattern discovery. This effectively creates an interdisciplinary “operating system” for science, allowing researchers to seamlessly invoke models and embed AI into scientific workflows.
Top scientific capabilities supporting frontline research
The “Panshi·Scientific Foundation Large Model” centers on core research needs, supporting scientific discovery with powerful, comprehensive expertise.
Its architecture adopts a heterogeneous mixture of experts, deeply customized based on open-source Chinese large models, integrating proprietary models for common scientific data modalities, and incorporating domain-specific models like AlphaFold and MatterGen.
In scientific ability, it systematically masters core theorems, laws, and knowledge across mathematics, physics, chemistry, materials, and biology, with top performance on international datasets and benchmarks like GAIA, SimpleQA, and the Human Ultimate Exam (HLE).
Figure: Scientific capability assessment
Figure: Complex reasoning problem assessment
Based on the “Panshi·Scientific Foundation Large Model,” the team also developed “Panshi Literature Compass” and “Panshi Tool Scheduler.”
“Panshi Literature Compass” helps researchers read articles, write reviews, and evaluate research topics and technical paths. It has accessed 170 million scientific papers and real-time open-source scientific information, capable of deeply understanding formulas and charts, and can review thousands of papers in 20 minutes, a task that previously took 3-5 days.
Figure: Panshi Literature Compass
“Panshi Tool Scheduler” aims to lower the barrier to using scientific tools, enabling autonomous planning and invocation of over 300 scientific computing tools, facilitating tool orchestration and quick setup of research workflows.
Figure: Panshi Tool Scheduler
Accelerating scientific discoveries across disciplines
“Panshi·Scientific Foundation Large Model” has been applied extensively in various fields, significantly speeding up research processes.
In life sciences, it supports the construction of X-Cell, a digital cell model that automates the entire process from gene sequences to cell phenotypes, aiding target discovery and virtual cell experiments, with efficiency over 10 times faster than traditional methods.
In high-energy physics, researchers rely on the model to automate particle physics workflows, improving simulation speed and reconstruction efficiency, aiding fundamental matter and universe research.
In mechanics, it predicts surface pressure fields for high-speed trains in various fluid environments, supporting design optimization.
Additionally, it assists chemists in synthesis, improves molecular structure predictions, and enables intelligent global resource scheduling for astronomical observations, continuously iterating to enhance practicality and reliability.
Application cases
The “Panshi·Scientific Foundation Large Model” has become an all-round scientific knowledge assistant, a super analyst of literature, and a professional problem solver, gradually evolving into a strategic advisor and think tank for scientific research, serving science and empowering research.
To promote “AI + Science” collaborative innovation, the Institute of Automation, Chinese Academy of Sciences, has launched the “Scientific Foundation Large Model Ecosystem Alliance” with over 40 research institutes, universities, and industry partners. They are actively promoting domestic computing power adaptation and building an open, autonomous, and controllable “AI + Science” ecosystem, contributing Chinese solutions to the global academic community, reshaping research paradigms, and opening up infinite possibilities in scientific research.
“Panshi·Scientific Foundation Large Model” is now fully open source. Experience it now!
Official website: https://scienceone.ia.ac.cn/