Domyn Small Arrives, LLM with 10 Billion Parameters Trained on CINECA's HPC Leonardo
The AI also speaks Italian. Domyn, a company based in Milan, has made available Dorym Small, a relatively limited parameter LLM (10 billion) supporting 50 languages, including English, French, German, Italian, Spanish, and Portuguese. This project was born from collaboration with CINECA, which provided the Leonardo supercomputer for training the model.
Dorym Small: 10 Billion Parameters for a Sovereign and Responsible AI Model
The AI race is mainly between the USA and China, powers investing hundreds of billions to innovate their models. Europe has been watching until now, risking not only being cut off from this sector but also having to rely technologically on foreign powers.
However, the situation is changing: even in the Old Continent, models are starting to be trained that can compete with the AIs from OpenAI, Anthropic, and others. Among them is Dorym, a company involved in the IT4LIA AI Factory initiative, which fits into the European AI Factory ecosystem. Dorym has made available the Dorym Small model, a smaller version (10 billion parameters) of Dorym Large, which is trained on as many as 260 billion parameters.
Dorym Small is designed to operate on systems with low computational power, such as notebooks and edge devices, making it ideal for on-premise use in all areas where low latency and data confidentiality are crucial. According to the company, the Domyn Small model offers performances comparable to those of competing models, while also ensuring a better balance between accuracy and efficiency, and it is able to compete with main models developed in the USA and China, positioning itself as a reference in its category in Europe. Benchmarks show that it can outperform in some tests Ministral-3-8B-Reasoning-2512, Llama-3.1-Nemotron-Nano-8B, and OLMo-3-7B-Think.
“Domyn Small demonstrates that Europe can lead in advanced AI by building open, sovereign, and production-ready reasoning models,” said Uljan Sharka, CEO of Domyn. “Through our established collaboration with CINECA, we are providing performance, efficiency, and complete governance to companies that need it most — those operating in regulated industries.”
CINECA, under the EuroHPC framework, contributes to the mission of putting sovereign and open high-performance computing at the service of industry, research, and public administration in Italy and Europe,” affirmed Francesco Ubertini, President of CINECA. “Domyn Small has been trained on a public European infrastructure, funded by government funds and hosted here in Italy. This is not a detail — it is the foundation upon which reliable and compliant European artificial intelligence is built, in accordance with the AI Act. CINECA's enabling role in this project demonstrates that public HPC infrastructure and industrial ambition in the AI field are not in competition but can operate synergistically for a European strategy on Artificial Intelligence that is both competitive and open.