Data sovereignty
Europe is introducing rules that data about Europeans must be controllable from Europe. For the data that matters, like client documents, accounts, and source code, where the systems live isn't a minor detail anymore.
· Deep AI research lab · Oslo, Norway
Unlocking intelligence
Tenki is a Norwegian deep AI research lab. We train and run our own language models on hardware in Norway, measure them against the work Norwegian companies actually do, and publish both the tasks and the results. What turns out to hold up, we bring into the companies.
Our own Norwegian model, run in Norway
Hugin and Munin are Norway's first publicly available local LLMs, named after Odin's two ravens. Underneath is our own Norwegian fine-tune, trained by us and run on hardware we own. Hugin takes the time it needs to reason; Munin answers in the moment. They are small and experimental, and we write down plainly where they fall short.
Read the model cards· Backdrop
Three forces, acting at the same time.
Three developments are acting on Norwegian companies at the same time. Over time, the combination decides who is left holding the value created.
Shadow AI
People already use ChatGPT, Copilot, and Gemini at work. Client documents, salary figures, customer lists pass through them every day. Not because anyone is careless, but because the tools work.
Competitive pressure
AI is a new infrastructure for work, much like the internet was in the '90s. The companies that didn't adapt to the internet largely don't exist anymore. That shift didn't come gradually either.
· Our answer
We build it, measure it, and run it ourselves.
Three parts, and the order is the point. We train the models ourselves, we measure them on Norwegian work with tasks that are public, and we run them in our own operations long enough to know where they fail.
Our own models
We train our own Norwegian fine-tune and run it on hardware we own, in Norway. It is free to try, with no sign-in.
Open measurement
We measure language models on Norwegian tasks and publish the tasks, the weighting, and the results. Including where we lose.
Our operations first
The systems run on our own operations before we mention them to anyone else. That is where we find out what does not actually work.
· Work with us
We take on a few engagements a year, with companies that want this inside their own operations. It funds the lab, and it is where we find out which problems are worth spending time on. If you are unsure whether it is for you, the maturity test is a reasonable place to start.
What we're thinking about.
Try it yourself.
The model is free and needs no sign-in. It is small, and it gets things wrong. We have written down where, and we measure it against the large models in open tests you can check.