Sevren Releases Embla, Its Most Efficient AI Model That Runs on a Phone
Copenhagen AI company Sevren has released Embla, which it calls the most efficient AI model ever built in Denmark. The model is named after Embla, the first woman in Norse mythology, because it is the first of its kind for the company.
Embla was trained from scratch and is small enough to run on an iPhone. It was trained on 1.5 trillion tokens using Gefion, Denmark's sovereign AI supercomputer, in what is described as the country's largest AI training project to date.
On performance, CEO Mads Aarup Pedersen says that in Sevren's own tests Embla delivers a better combination of speed and quality than comparable models from Google, IBM and Liquid AI. These are the company's internal results, not independent benchmarks.
The release follows a recent partnership between Sevren and DCAI, which gave the company access to Gefion, including NVIDIA DGX B300 systems powered by Blackwell Ultra GPUs. The team was live on the infrastructure less than a week after signing the agreement. Embla is the first public release to come out of that work.
The release matters as an example of a Danish company using the country's national AI infrastructure to train a model of its own, and of a compact model designed to run directly on consumer hardware rather than in a data centre.