Publication date: October 19, 2026
This is one of the most important premieres in recent months in the world of Polish artificial intelligence. Created as part of Gaia AI Factory, the Polish ModernBERT family of language models sets a new standard for understanding the Polish language and already demonstrates results surpassing previous comparable Polish language models. Importantly, this is not another chatbot. It is a foundation on which modern, cost-effective AI systems for business, administration, and science can be built.
A Polish breakthrough in AI
In the world of artificial intelligence, the most attention is drawn to generative models that can write texts or answer questions. However, behind many successful business solutions stand less visible yet equally important technologies. Polish ModernBERT belongs precisely to this group. The model was designed for one main task: the best possible understanding of text in the Polish language. As a result, it can form the basis of systems that analyze documents, classify content, search for information, or recognize key concepts in enormous datasets.
As emphasized by Dr. Eng. Marek Kozłowski, head of AI LAB at the National Information Processing Institute (Ośrodek Przetwarzania Informacji – Państwowy Instytut Badawczy), the center that built these models as part of Gaia AI Factory:
Polish ModernBERT is a representation model whose task is high-quality understanding of text in the Polish language and building its deep representation. It is not intended for generating content, but for analyzing, comparing, or even searching for information.
/p
The best result in the history of Polish models of this type, i.e. representation (encoder-only) models
In studies covering as many as 30 diverse tasks related to natural language understanding, Polish ModernBERT achieved the best results among the compared analogous Polish models. It achieved particularly impressive results when working with long documents, such as legal acts, reports, or extensive studies.
The model can analyze documents of up to 8,000 tokens, maintaining high effectiveness while consuming fewer computational resources than earlier solutions. In the Base variant (capable of efficient operation even on CPU), it achieved better results while reducing the number of parameters by 22% compared to the reference model used for comparison.
Millions of documents? That is exactly its environment!
The greatest potential of Polish ModernBERT is revealed wherever organizations must work with enormous collections of documents, even long ones often spanning several or a dozen or so pages. The model can be used for:
- automatic document classification,
- assigning tickets to appropriate categories,
- detecting proper names, such as names of institutions, people, and places,
- identifying semantically similar content,
- grouping documents by similar topic,
- building modern semantic search engines,
- creating RAG systems supporting generative AI.
- building fast, training-free models of the System-1 type
This means real applications in public administration, banking, the legal sector, healthcare, industry, and science. Everywhere that tens of thousands or millions of text documents are processed every day. It should be remembered that most work concerning the automation of document processing based on text analysis does not always require the use of the most expensive generative models; in most tasks, it is enough to use representation models so that the business benefit is quickly felt and costs remain marginal.
AI available to all companies
One of the greatest advantages of Polish ModernBERT is its efficiency. The new model is not only more accurate than earlier solutions, but also faster and more economical in terms of the hardware resources it consumes, or indirectly even its energy use. It can run even on standard CPU processors, without the need for expensive GPU cards. This is good news for many institutions and companies that want to implement artificial intelligence but do not have extensive computing infrastructure equipped with GPU clusters.
As Marek Kozłowski emphasizes:Our models are much faster than earlier solutions and significantly smaller while maintaining high quality. Even organizations that do not have GPUs can use them. The models are still efficient and can process several texts per second even on classic CPU units.
Created in Poland, developed for Europe
The Polish ModernBERT family was trained on a purpose-built corpus containing over several dozen billion Polish tokens and specially adapted to the specifics of the Polish language. This makes the model better at handling inflection, declension, and the complex structure of our language than many similar multilingual solutions. The creators, however, do not intend to stop at Poland. The next step may be expanding the technology to more European languages, which could open the way to building a new generation of European AI models based on local data, languages, and user needs.
A foundation for European artificial intelligence
Polish ModernBERT shows that breakthrough AI technologies can also emerge in Europe and be developed with specific user needs in mind. It is not another chatbot, but a key element of core, foundational infrastructure that will make it possible to build more intelligent, faster, and cheaper information processing systems. For companies, public institutions, and organizations with enormous document resources, this means one thing: artificial intelligence is beginning to truly understand the Polish language, and it does not always have to require expensive investments in GPU cards.
And this is only the beginning. Polish ModernBERT, created as part of Gaia AI Factory, may become one of the most important foundations for the development of Polish and European AI in the coming years.
Polish ModernBert can be used here: https://huggingface.co/collections/OPI-PIB/pl-modernbert
Gaia AI Factory – this is the Polish Artificial Intelligence Factory being created as part of the EuroHPC JU program. It combines modern supercomputing infrastructure, data, competencies, and expert support to accelerate the development of innovative AI solutions for science, business, and the public sector. It is one of the largest artificial intelligence infrastructure projects currently being implemented in Poland. The consortium is formed by eleven top Polish
