Domyn Edge: Italy's open 4B-parameter model is being released piece by piece
Italian AI company Domyn began a phased release of Domyn Edge on October 20, 2026. It's a compact agentic language model with 4 billion parameters, trained on the Leonardo supercomputer. Intermediate checkpoints and code are out, but the finished model is still missing.
What this is about
Italian AI company Domyn (known as iGenius until June 2025, founded and led by Uljan Sharka) began a phased release of Domyn Edge on October 20, 2026. It is a compact "agentic" language model with 4 billion parameters. Parameters are, roughly speaking, the adjustable values a model learns during training. "Agentic" means the model is meant to carry out sequences of actions, not just answer single prompts. According to the main source (tech-insider.org), all details come from Domyn itself and are current as of October 30, 2026.
What's available and what's missing
Domyn isn't doing a classic one-shot release. Instead, it is sharing intermediate stages from training. Available now:
- the pre-trained base model,
- intermediate checkpoints (saved training states) from pre-training, mid-training and supervised fine-tuning,
- a "student" checkpoint, which usually means a model produced by distillation (learning from a larger model),
- the training code on GitHub. The model itself is on Hugging Face.
Still to come are the final post-trained model, the full datasets and further research materials. There is no version label such as "Edge 1.0". Domyn has announced more post-training phases through November 2026. According to the source, the license isn't definitively confirmed yet. Domyn talks about "open licenses". There is no price list or API.
Training and numbers
| Feature | Detail |
|---|---|
| Parameters | 4B |
| Training data | about 13 trillion tokens, from scratch |
| Machine | Leonardo (CINECA Bologna, EuroHPC) |
| GPU hours | about 2.3M for pre-training, about 5M in total |
| Context | trained up to 1M tokens, 256,000 recommended |
13 trillion tokens is a lot for a model this small ("overtraining"). It's a common way to make small models strong, so they fit on a single graphics card or a company server. According to Domyn, Edge can run locally or on-premises, including in air-gapped environments. That targets banks, government agencies and defense. According to the source, no benchmarks are available, including for long context. A token is a chunk of text the model works with. Context is the amount of text it can keep "in view" at once.
Edge is not EUROPA
The source stresses the risk of mix-ups. EUROPA is a separate, much larger project. On June 19, 2026, the EU Commission picked the Domyn-led consortium as winner of the Frontier AI Grand Challenge. It is meant to build an open model for all 24 official EU languages, with more than 400 billion parameters, and gets up to 2.5% of EuroHPC compute capacity for one year. Press reports mention around 500 billion parameters, which the source says is not official. Domyn gives the second half of 2027 as the timeframe. According to Atalayar, Henna Virkkunen, Executive Vice-President of the EU Commission, said Europe can lead in advanced AI on its own terms. Whether Edge covers all 24 languages isn't documented.
Background
Domyn has made several moves in 2026. Domyn Small (10B parameters, under an MIT license according to Domyn) arrived in May, a partnership with Cognizant followed in July, and Domyn Large debuted on Microsoft Foundry in August. On October 15, Domyn also published a paper on specializing NVIDIA Nemotron 3.5 Super VL. On 353 test passages, the share of facts found in a single pass rose from 16% to 33%, and rule violations fell from 8.7% to 3.8%. These are vendor figures for a very specific task.
What you need to try it
A 4B-parameter model runs quantized (at reduced numerical precision) on a graphics card with about 4–6 GB of VRAM (video memory), on Apple Silicon Macs, or, if need be, on a CPU with 8–16 GB of RAM. Plan for a few GB of storage. Long context takes extra memory. The source doesn't say whether ready-made GGUF files or an Ollama entry exist. The available versions are base and intermediate models, not finished chat assistants.
Our take
Edge's appeal lies less in performance than in traceability. Intermediate checkpoints, code and announced data allow audits, which matters for regulated industries and the EU AI Act. Domyn's claim of AI Act compliance hasn't been independently verified. The license, final model, benchmarks and pricing are all still open. A 4B model won't replace frontier models on complex tasks. For now, treat Edge as a base for experiments and research, not a finished product.
Sources
- Google News (IT) tech-insider.org
- atalayar.com atalayar.com
Try it yourself: Domyn Edge
Explained step by step, no prior knowledge needed – on your own computer or free in your browser.
What can I do with it?
Domyn Edge is a small language model with 4 billion parameters, built for agentic tasks, meaning multi-step workflows that use tools. You can run it locally, even without an internet connection. So far the base model, intermediate checkpoints and a student checkpoint are available. The finished post-trained model is due by November. For now, it's more a research and experimentation model than a chat assistant. No benchmarks exist yet, so you can't reliably compare its quality with well-known small models. Its strength is transparency: you can trace how the model was built. A model this small remains limited on complex tasks.
Who is it worth it for?
Of interest to developers, researchers and companies who want to test a small, traceable model locally or adapt it to their own data. If you want a finished, well-tested assistant right now, wait for the final model in November.
Ideas: What you can do with it
Have the model sort incoming mail or tickets into categories locally. It works without the cloud and is a good test for a small model.
Pull amounts, names and dates out of internal texts and output them as a list. It makes a good first real-world test with sensitive data that never leaves your machine.
Compare intermediate checkpoints to see how behavior changes between pre-training and fine-tuning. This is especially interesting for teaching and research.
Use the base model as a starting point for fine-tuning on your own domain texts. With 4B parameters, that's doable on a single graphics card.
What your computer needs
Values estimated from the model size – the Model Card may have more precise figures.
Which variant is right for me?
Many models come in several sizes or "quantizations." These are compressed versions: smaller and faster, but slightly less accurate. Rule of thumb: pick the largest variant whose memory requirement fits in your VRAM (video memory). The scripts below choose this for you automatically.
| Variant | Download | Needs approx. | Compatible graphics cards |
|---|---|---|---|
4-bit (Q4) Default |
2.3 GB | 4 GB | almost any graphics card with 4 GB or more (e.g. GTX 1650, RTX 3050) |
8-bit (Q8) |
4.2 GB | 6 GB | e.g. RTX 2060, RTX 3050 8 GB, RX 6600 |
16-bit (Original) |
7.8 GB | 10 GB | e.g. RTX 3060 12 GB, RTX 4070, RTX 5070, RX 7700 XT |
The base model, intermediate checkpoints and a student checkpoint are available. For trying it out, pick the most heavily trained version on offer. The source doesn't mention quantized versions. If you're a beginner, it's worth waiting for the final post-trained model.
- At 4B parameters, roughly 4–6 GB of VRAM is enough when quantized, and about 8–10 GB unquantized (16-bit).
- Running on a CPU with 8–16 GB of RAM is possible, but slow.
- Apple Silicon Macs with 16 GB are well suited for testing.
- The recommended context of 256,000 tokens needs far more memory than short prompts, so start with small contexts.
- Google Colab with a free GPU should be enough for a 4B model.
How do I find out what my computer has?
- Windows: Press Ctrl + Shift + Esc → Task Manager → Performance tab → click GPU on the left. At the bottom you'll see Dedicated GPU memory (that's your VRAM). You'll find system memory under Memory.
- Mac: Apple menu at the top left → About This Mac. You'll see your chip (e.g. M2) and memory (e.g. 16 GB). On Apple chips, graphics and processor share this memory.
- Linux: Type
nvidia-smiin the terminal (shows your graphics card and its memory) andfree -hfor system memory.
Guide
Google Colab (in your browser, no installation)
Google Colab is a free service from Google. It gives you a computer with a graphics card in your browser for a few hours. All you need is a Google account. For this model: runs for free – The free Tesla T4 GPU with 15 GB is enough.
- Download the notebook. Click the button and save the file.
- Open Colab. Go to colab.research.google.com and sign in with your Google account.
- Upload the notebook. In the window that appears, click Upload on the left (or choose File → Upload notebook at the top) and select the file
colab.ipynb. - Turn on the GPU. In the top menu, go to Runtime → Change runtime type → under Hardware accelerator, select T4 GPU → Save.
- Start everything. Runtime → Run all (keyboard shortcut Ctrl + F9). If you see the warning “This notebook was not authored by Google,” click Run anyway.
- Wait. The cells run one after another. A green checkmark appears to the left of each cell when it's done. Installation and download take about 5 minutes here.
- Try it out. The result appears below the last cell. To test something else, change the text in the cell and run it again with Shift + Enter.
- Clean up. When you're done: Runtime → Disconnect and delete runtime. This saves your free quota.
Good to know: The free GPU is limited (usually a few hours per day). After about 90 minutes of inactivity, Colab disconnects, and you'll have to start everything over.
Windows 10 / 11
This model runs on Python. Our script handles the setup automatically: it creates an isolated Python environment, installs the required packages (PyTorch, approx. 3 GB) and downloads the model (2.3 GB (about 3 min at 100 Mbit/s)).
- Update your graphics driver (NVIDIA): install the latest driver. Without an NVIDIA card, it runs on the CPU – much more slowly.
- Install Python. Download Python 3.12 from python.org and run the installer. Important: At the bottom of the first window, tick Add python.exe to PATH, then click Install Now.
- Create a folder and download the files. Create a folder, e.g.
Documents\AI-Test, and save both files into it: - Open a terminal in the folder. Open the folder in File Explorer, right-click an empty spot → Open in Terminal.
- Run it. Paste this command and press Enter:
Installation and download are running now. The first time, this can take 10–30 minutes.powershell -ExecutionPolicy Bypass -File try_windows.ps1 - Result. The window shows the model's answer to a test question.
- Try your own input: Open
demo.pyin an editor, change the sample text, and save. Then run this in the terminal:.\.venv-domyn-edge\Scripts\Activate.ps1 python demo.py
macOS
Macs with an Apple chip (M1 to M5) are ideal. Older Intel Macs are very slow.
- Open Terminal. Press Cmd + Space, type Terminal, and press Enter.
- Install the developer tools (first time only; includes Python):
Click Install in the window. If you see "already installed," everything is already in place.xcode-select --install - Download the files – both end up in your Downloads folder:
- Run it:
The first time, installation and download take 10–30 minutes.cd ~/Downloads bash try_mac.sh - Result. The window shows the model's answer to a test question.
Linux
- Check your GPU.
nvidia-smishould show your card. If it doesn't, install the driver on Ubuntu and restart:sudo ubuntu-drivers install - Install the Python basics (Ubuntu/Debian):
sudo apt update && sudo apt install -y python3-venv python3-pip git - Put the files in a folder and open a terminal there:
- Run it:
bash try_linux.sh - Result. The window shows the model's answer to a test question.
Try it
Summarize the following paragraph in two sentences: …Classify this email as invoice, advertising or support request, and briefly explain why.Extract all dates and amounts from this text as a list.Plan, step by step, how you would calculate monthly revenue from a CSV file.Explain the difference between a base model and a post-trained model in simple terms.Tips for better results
- Start with short, clearly defined tasks such as classification or extraction.
- Note exactly which checkpoint you use, since versions change.
- Compare results with a model you already know.
- For now, use much shorter contexts than 256,000 tokens.
If something goes wrong
The final model hasn't been released yet. The available files are base and intermediate versions, which don't necessarily answer like a chat assistant.
The final license is unconfirmed, so check it before any commercial use.
The source doesn't confirm ready-made GGUF files or an Ollama entry. You may have to convert the model yourself or load it with Transformers.
Don't confuse Edge with EUROPA, the later model with 400B+ parameters.
Support for all 24 EU languages isn't confirmed, so test German yourself.
“out of memory,” a crash, or your computer freezes
The model is too big for your VRAM (video memory). Pick a smaller variant from the table above and close other programs, especially games and browsers with lots of tabs.
It runs, but extremely slowly
Your CPU is probably doing the work instead of your graphics card. In Ollama, the PROCESSOR column of ollama ps shows how much runs on the GPU. If it says "100% CPU", update your graphics driver or pick a smaller variant that fits entirely into VRAM (video memory).
“ollama” or “python” is not found
Close the terminal window and open it again so it recognizes the newly installed program. For Python on Windows: reinstall it and check Add python.exe to PATH.
Windows: "Running scripts is disabled on this system"
Run the script with exactly the command above: powershell -ExecutionPolicy Bypass -File try_windows.ps1. This allows execution for this one time only.
Error message when loading the model (version, "trust_remote_code", unknown architecture)
Brand-new models often need the latest software version. With Ollama: update Ollama. With Python: pip install -U transformers in the activated environment. The Model Card usually lists the exact version.
Tokenstrom generates the scripts and notebook automatically. Please take a quick look before running them – they're plain, readable text. Mind the model's license: apache-2.0.
Ask the AI assistant
Questions about this article, installation or background? Our assistant knows this page and runs on our own machine, not in someone else's cloud.
The Assistant is an AI (Gemma 4) and can make mistakes – please check important details against the source. Questions are used only to generate an answer and are deleted after one hour. Please don't enter any personal data.
Comments
What do you think? Already tried it? Write it here, no sign-up needed.
Comments are public. Links aren't allowed, and we delete abusive posts. You can delete your own comment at any time in this browser. Learn more in the Privacy Policy.