April 17, 2026
Numbers are the language of science, but in research papers they are often buried in text and difficult to analyze. Jülich researchers have developed an AI system that automatically identifies, categorizes, and transforms these numbers into structured data. Therefore, the Quinex framework eliminates the need for time-consuming manual tasks.
Whether in energy, climate, or materials research, scientific papers are full of numbers, or more precisely quantitative data, such as efficiency, temperature, cost, and emissions. These are often important for improving models and identifying trends. At the same time, the number of scientific publications is also rapidly increasing. For many research questions, it is currently virtually impossible to manually evaluate all relevant publications, and the time and resources required would be enormous.
The Quinex (“Quantitative Information Extraction”) framework developed by Jülich researchers is based on language models and automates this process. Artificial intelligence identifies numbers, assigns them to the appropriate units, and recognizes what, when, where, and how it was measured. Thus, a statement like “Efficiency levels in 2025 are assumed to be between 63 and 71 percent” is transformed into a structured dataset containing all relevant contextual information, from year and measurement method to source.
Open and efficient AI
Unlike many proprietary AI solutions, Quinex is completely open and based on a relatively small and efficient language model. They are specially trained to recognize and classify quantitative information within scientific documents. Compared to similar systems, Quinex provides more accurate results, captures contextual information in a more nuanced way, and also takes into account implicit characteristics.
Despite its compact size, Quinex achieves approximately 98 percent recognition accuracy (F1) for numbers and related units, and approximately 87 percent and 82 percent for quantified properties and entity classification. These high accuracies were achieved through specially created training datasets and methodological improvements.
“We wanted to develop a powerful yet transparent and resource-efficient tool,” explains Dr. Jann Weinand, Head of Integrated Scenarios at Jülich System Analysis. “Quinex makes artificial intelligence more accessible for data analysis in science.”
Passed practical exam
To test Quinex’s practical suitability, the system was applied to thousands of scientific abstracts from various fields. It successfully extracted data on power generation costs for different energy technologies, maximum human oxygen uptake, magnitude and location of earthquakes, and band gaps of photovoltaic materials.
Information can be automatically checked and referenced. Changing times: A detailed analysis to understand Quinex’s characteristics and trends.
A new perspective for Forschung
“Sprachmodelle eröffnen neue Perspectiven für die Wissenschaft und helfen dabei, den Überblick über ganze Forschungsbereiche zu behalten”, sagt Hauptautor Jan Göpfert. “Sie ermöglichen automatisierte Literaturrecherchen, den Aufbau einheitlich strukturierter Forschungsdatenbanken und Trendanalysen, die Entwicklungen in Wissenschaft und Technik frühzeitig sichtbar machen.”
“Unser Ziel ist es, Forschende von Routinearbeit zu entlasten,” says Dr. Patrick Kukerz of the Reiter der Group Forschungs Administration. “Quinex in Helfen, Schneller in Erkentnissen, Gelangen, and Waxende in Deitenfurt, Wissenschaft in Beherschen.”
Grenzen and Kunftige Verbesselungen
Quinex is not completely error-free either, but transparency is part of the design.
“The system recognizes numbers and units very reliably,” says Jan Göpfert. “These are taken directly from the text, so there are no ‘hallucinations’. However, misunderstandings can sometimes occur, such as when important references are scattered throughout the text. ”
Therefore, Quinex remains a tool to support people, but not to replace them. “We recommend using Quinex to inform and reassure researchers, but the responsibility for interpreting the results lies with the researchers,” says Göpfert. All recognized numbers can be traced back to their source and are highlighted with the original text where possible.
The team is working to further develop Quinex by adding domain-specific datasets and models to make it even more efficient and flexible to adapt to different research requirements.
Welcome to open collaboration
Forschungszentrum Jülich is making Quinex available as an open source project.
It aims to give researchers around the world the opportunity to test, extend and adapt the system to their fields, from energy research to chemistry to biomedicine.
Quinex open source: https://go.fzj.de/quinex
Jan Göpfert, Patrick Kuckertz, Gian Müller, Luna Lütz, Celine Körner, Hang Khuat, Detlef Stolten, Jann M. Weinand (2026): Quinex: Quantitative information extraction from text using an open and lightweight LLM. Innovation. DOI: 10.1016/j.xinn.2026.101391
/Open to the public. This material from the original organization/author may be of a contemporary nature and has been edited for clarity, style, and length. Mirage.News does not take any institutional position or position, and all views, positions, and conclusions expressed herein are those of the authors alone. Read the full text here.
