Domyn begins rolling out Edge, sharing the model and its training recipes to unlock a new generation of sovereign AI
- Open development, shared opportunity: built on public compute, Domyn Edge’s training checkpoints, recipes and tools help researchers and builders study, adapt and create their own AI.
- The best-performing fully open model of its size: published benchmarks place Edge ahead of open models in its category.
- Compact and cost-effective: the 4-billion-parameter agentic model offers speed and simpler, cheaper specialization than larger models.
- Local deployment, institutional control: Edge runs on compatible local hardware, bringing AI closer to operations with more choice over where it runs.
The rolling release starts with incremental checkpoints obtained during pre-training, mid-training and supervised fine-tuning, and will progress along different post-training stages into November 2026.
Trained from scratch on CINECA’s Leonardo supercomputer, Edge extends the value of European public compute to the wider AI community. By sharing the model at intermediate stages of training and the resources to develop it further, Domyn is opening a path for others to create new capabilities beyond its own final release. The goal is to turn the development of one model into the foundations for many more.
Domyn is bringing Edge to this year’s NVIDIA GTC in Berlin, highlighting how open, efficient models can enable organizations to develop specialist intelligence and run it under their own control.
Luca Antiga, CTO of Domyn, said “Sovereignty is about control. Domyn’s mission is providing organizations with the control they need for AI to pervasively power their processes. Today we are happy to kick-off the roll-out of Edge with the initial batch of model checkpoints, recipes and its open source tooling. Researchers can explore how capabilities develop, builders can choose a starting point and continue training toward their own goals. Europe needs an ecosystem of builders, and the real potential of Edge is in what the ecosystem will be able to create, together with us or independently: new capabilities and specialist intelligence they can fully control.”
Incremental roll-out, shared opportunity
Intermediate checkpoints are saved versions of a model at different stages of training. Releasing them gives researchers visibility into how Edge evolves and technical teams a choice of starting points for their own development.
With the accompanying recipes and tools, builders can shape subsequent training around their requirements, taking the model in a different direction from Domyn’s final version. A bank, for example, could adapt Edge around its terminology and transaction-monitoring processes, through its own AI team or in collaboration with Domyn.
This approach connects openness with practical sovereignty: the freedom to develop intelligence around an institution’s knowledge, choose where it runs and shape how it evolves.
Making specialist AI practical
Domyn Edge is designed for multi-step agentic work and tool use. Its compact size reduces the compute needed to adapt and run it, making specialist intelligence more practical and cost-effective to develop.
Once adapted to a specific task, it can provide an efficient alternative to a larger, general-purpose model. It can also work alongside larger models within an agentic system, with each handling the tasks it is best suited to.
Local deployment on compatible hardware brings that intelligence into operational settings such as manufacturing, where substantial GPU infrastructure would be impractical. Institutions can run AI closer to the work, within environments they control.
Availability
Today, the staged release is starting on Hugging Face, alongside the open source release of training code on GitHub. Details of checkpoints, training recipes, tools, supported hardware and the final release roadmap will be published here.
Find out more about Domyn Edge in the dedicated paper.
About Domyn
Domyn builds sovereign AI for regulated industries, giving organizations greater ownership and control of the intelligence they rely on. Its integrated operating system combines proprietary models with enterprise knowledge, industry-specific agents, governance and infrastructure, enabling organizations to build AI around their business, adapt it to their needs and trust it in mission-critical environments.
Domyn works with major enterprises and institutions, including the Italian government, Fincantieri and BNY, which is also an investor. Eurizon Capital, Rabobank and Abu Dhabi's G42 are among its other backers.
Domyn is advancing large-scale European AI infrastructure through Colosseum, Europe’s largest AI supercomputer, developed in partnership with NVIDIA and G42. In June 2026, the European Commission selected EUROPA, a European consortium led by Domyn, as the winner of the Frontier AI Grand Challenge to develop an open-source artificial intelligence model covering all 24 official EU languages.
For more information, visit domyn.com.
