Zhejiang University Unveils Qiushi Engine with Groundbreaking Long-horizon Autonomous Scientific Research Capabilities

According to official releases, a new artificial intelligence system named Qiushi Engine has been officially launched to deliver independent and continuous scientific research capabilities. Developed by a research team from the College of Information Science and Electronic Engineering at Zhejiang University, the platform achieves industry-leading thousand-step long-horizon scientific reasoning, enabling autonomous organisation and advancement of comprehensive research workflows to deliver proactive technical support for global scientific exploration.

Traditional large language model research tools have long been limited to auxiliary functions, including literature retrieval, code writing, data processing and automated procedural operations. These tools rely heavily on researchers to define precise questions and only optimise partial research links, lacking the capacity to tackle open-ended scientific problems with uncertain research paths. Independent judgment, dynamic adjustment and sustained exploration of original scientific discoveries have remained a global technical bottleneck for intelligent research systems.

Qiushi Engine breaks the inherent limitations of existing intelligent research tools by focusing on full-process scientific research iteration. The system can decompose overall research goals into executable tasks, transform innovative ideas into verifiable solutions, assess the credibility of preliminary findings, and revise research schemes autonomously when facing suboptimal results. It operates through continuous trial, error correction and re-verification to promote in-depth exploration of open scientific topics, realizing stable thousand-step long-horizon reasoning cycles.

The system adopts a multi-agent collaborative framework to simulate the operational mode of professional research teams. Multiple specialised agents undertake divided responsibilities for research route planning, methodological construction, task execution, result analysis and risk inspection. The collaborative mechanism supports the steady advancement of complex scientific research tasks, enabling the AI platform to operate as a new-type autonomous research tool rather than a simple auxiliary service.

Practical optical experiment verification has proven the system’s robust autonomous research capabilities. With only an open research objective input, Qiushi Engine can operate continuously for more than ten hours to complete full-link research procedures, including literature investigation, theoretical analysis, experimental scheme design, programming, data analysis and result evaluation. By iteratively troubleshooting errors, revising schemes and verifying outputs based on real experimental feedback, the system has independently produced multiple original research findings with verifiable conclusions.

The intelligent platform drastically improves scientific research efficiency. It can trigger thousands of large model calls within hours or dozens of hours to complete massive intermediate exploration work, while equivalent manual research procedures usually require weeks or months of concentrated effort from professional researchers.

Different from conventional AI models that only generate textual content and code, Qiushi Engine is specially designed to adapt to real and complex scientific research scenarios. It can interact with computing platforms, data analysis tools, experimental equipment and professional research software, accurately identifying tool functional boundaries and dynamically adjusting research strategies according to multi-dimensional operational feedback. The design avoids hollow theoretical deduction and ensures all research outputs match practical experimental conditions.

The launch of Qiushi Engine marks a shift in artificial intelligence application from passive question answering and partial task execution to active whole-process scientific exploration under human supervision. It realizes standardized management of goal setting, process promotion and result review throughout scientific research activities.

Global technological competition is shifting from large model performance iteration to practical complex problem-solving capabilities in real scenarios, with basic scientific research standing as a core competitive track. At present, Qiushi Engine has carried out autonomous research across more than ten disciplines including physics, optics, life medicine and mathematics. It has proposed innovative theoretical methods for long-standing fundamental problems in computational physics, established new basic theoretical frameworks in spectroscopy, and discovered original optical computing mechanisms. All innovative outcomes have passed internal multi-round verification and are scheduled for public release in due course.