Jan 12, 2026

Poles Have Their Own LLM

If you ask people what AI models they know, you can expect to hear models offered by the largest providers such as ChatGPT, Gemini, or Claude. In this article, I present two, in my opinion, most important initiatives led by Poles. These are PLLuM and Bielik. The goal of these projects is to create models that better understand Polish, know our Polish cultural customs and the specifics of local applications, and be generally available to Polish society.

PLLuM – Polish Large Language Universal Model

The largest project in Poland is PLLuM. It is a research and development initiative aimed at creating a large, open language model for Polish. The project was created thanks to a consortium of Polish scientific and research institutions, including the Wrocław University of Technology, NASK, OPI PIB, PAN, University of Łódź, and its funding comes from the Ministry of Digitization. The goal of PLLuM is to create a model that will:

be available under an open license, with the possibility of use in administration, business, and research,
be adapted to Polish language and Polish reality,
be used, among other things, as an intelligent assistant supporting citizen service in public services.

The project has been developed since early 2024. The completion of work and the provision of first results took place in 2025. PLLuM models are already being made available in various variants and tested in practice, including in a pilot program in public administration. It is currently used in the mObywatel application for user support. ¹

An example of PLLuM application is also a pilot program in the city of Częstochowa, where the model supports the automation of local administrative services – from writing official documents to analyzing resident inquiries. ²

PLLuM is also intended to be an element of building Poland’s technological sovereignty, i.e., independence from foreign AI model providers. You can read more about PLLuM here

Bielik – A Polish Open Source Model

Another important Polish LLM project is Bielik AI, developed by the SpeakLeash Foundation in cooperation with the Academic Computer Center Cyfronet AGH. It is a family of language models designed with Polish language in mind – including a large model called Bielik-11B-v3, which has about 11 billion parameters.

Bielik stands out for:

a large, proprietary database of Polish language texts,
open code and accessibility for the AI community,
various model variants of different parameter scales (e.g., smaller versions 1.5B and 4.5B in newer iterations).

Newer versions of Bielik achieve good results in international benchmarks comparing language models, indicating their competitiveness against other LLMs in the world.

The Foundation runs several interesting projects. Ordinary citizens are encouraged to participate in these projects. Our involvement is meant to help and accelerate the development of subsequent models. These projects are:

Citizen Bielik

This is a project aimed at encouraging Poles to develop subsequent models from the Bielik family. An application has been prepared to make it easier to send data and describe it. This data will be used to train subsequent versions of the Bielik model. More about this initiative can be found at https://obywatel.bielik.ai/

Sójka AI

Another project run by the SpeakLeash Foundation. Its purpose is to detect harmful content in real-time. In my opinion, this is an incredibly important project at a time when the language used in the media is at a low level. I encourage you to learn more at https://guard.bielik.ai/

Other Initiatives and Projects

Besides PLLuM and Bielik, there are several smaller or more specialized projects related to LLMs in Poland:

Research projects on language resources and infrastructure, e.g. as part of CLARIN-PL, which aim to support the development of NLP tools and future language models adapted to the Polish language. ³

Industry experiments with language models for specific sectors, e.g. a project by OPI PIB and PKO BP focused on models for applications in finance and industry data analysis. ⁴

Community and experimental ventures, such as informal proposals for local models (e.g., Polka 1.1B or various fine-tunes of base models) published and developed in open source communities.⁵

Why Polish LLMs Matter

Polish LLM models are built primarily with a view to:

better understanding and generating Polish language, taking into account cultural and linguistic nuances,
technological sovereignty – less dependence on foreign AI providers,
practical applications, such as chatbots, automation of administrative processes, text analysis, and support for expert work.

Compared to foreign LLMs, Polish models can offer better adaptation to local language and legal needs as well as greater control over the data used in the training process and applications.

How to Use Polish LLMs?

You can use both models the same way as popular models through chat. Remember that these models are not as large as the models of the “giants.” You need to write well-defined prompts to get the expected results. I wrote about how to write good prompts in this post.

https://chat.bielik.ai/

https://pllum.clarin-pl.eu/pllum_8x7b

It is also possible to run models locally. But this requires some technical knowledge and an appropriately powerful GPU with adequate RAM. I use ollama. However, due to hardware limitations, the speed of generated tokens is low. Nevertheless, it is sufficient for simple experiments and work.

Summary

The Polish LLM scene is developing dynamically. Major projects, such as PLLuM and Bielik, achieve results that can truly compete with international solutions in specific language tasks. Their development is supported both by public and scientific institutions and by the open source community. I hope that these investments in LLMs are a step towards our technological independence and better use of AI in public and commercial applications.