Our research
In Sevren, we believe in transparency. We would rather be challenged in public than admired quietly.
Sevren is a European AI research lab. We are developing efficient LLM architectures from the ground up, doing more with less.
Built in Europe. Open by design. Sovereign by construction.
AI is available to everyone, but built by a few
Today, Europe stands at a crossroads. The AI that shapes our society and our economy is built in the US and China, and we must decide whether to remain dependent on their technology or develop AI on our own terms.
At Sevren, we are making that decision easier. We are building the foundation that enables European industry and academia to develop and control their own AI.
We are paving the way for European AI sovereignty. We have developed new ways to reengineer the world’s leading open-source models, making them more efficient while substantially reducing the computing power and energy they require.
We own what we build, and we document our methods openly, so that anyone can use them – including our competitors.
Together, Europe can develop the AI it needs without having to outspend the US and China. Because we can prove that technological progress is not about who builds the largest models first. It is about who builds the smartest.
In Sevren, we believe in transparency. We would rather be challenged in public than admired quietly.
Europe competes best on efficiency and openness, we build AI that needs less resources and share what we learn.
Sovereign by construction, on the way to sovereign by origin.
A company's know-how tends to become everyone else's commodity if put into a closed model the company does not own.
The destination is sovereign by origin, with models that start here rather than rebuilds of somebody else's models.
Bridging research and industry in the open.
AI has reached everyone, but is built by a few. Sevren works to change who gets to shape it going forward by bridging the international research frontier with Europe's own companies.
We do this by documenting methods openly and collaborating across borders – competitors included.
Weights, research notes and reference code ship in the open.
Every watt saved buys another experiment.
The cost of intelligence is, in the end, an energy bill. Sevren rebuilds existing models so they need less compute and less power to do the same work.
Efficiency is not thrift. Efficiency is speed.
Every watt, every gigabyte of memory and every hour cut from a run buys the research loop another turn. The best AI models are built by those who can conduct 100 experiments the fastest.
Open models are a starting point, not a finish line. Sevren exists to take the best of already existing models apart, rebuild the parts that matter and keep Europe's most valuable knowledge in its own hands.
Sevren is a European AI lab, based in Copenhagen, developing open-source AI technologies that make advanced AI more computing-efficient, through Faberon, our research harness, and HEIMR, our research programme.
To develop foundational AI technologies that make advanced AI significantly more efficient, while helping advance the wider AI ecosystem through open collaboration: Sovereign by construction today. Sovereign by origin next.
Our research programme on building, training and serving models for phones and laptops.
Language models are naturally sparse. Most of what they compute contributes nothing to the answer, yet the standard recipe pays for it anyway, in compute, memory and energy.
We design models that avoid spending compute on work that makes no difference to the output. Around those designs we build the training methods and kernels that turn the design into measured gains rather than theoretical ones.
Sevren’s controlled recursive self-improvement engine.
It proposes, builds and measures new research directions and lets what works shape the next attempt – most attempts are intentional failures, which is how the loop learns.
The lab does not keep office hours: Faberon does not stop when we go home for the day.
Built to support a cooperative European open-source strategy, it shares methods and results across the community and helps offset the limited resources of Europe’s sovereign datacenters.
In Sevren, we believe the best person for the job is not a compromise on team fit, but a condition of it. We would rather stay lean and sharp than grow and dilute.
We offer the next building stone in your career. You offer the craft to shape it.
If you don't see a fit, but think you should be here anyway, write to us at hello@sevren.ai.
