We are excited to share a major milestone for TerraNex AB.
TeraNex has developed and benchmarked the first version of a multilingual tokenizer, a critical component in their work to build a sovereign European foundational AI model.
Performance highlights: Early results show performance on par with leading systems like DeepSeek and GPT in coverage, and surpasses both in efficiency. All development and benchmarking were carried out on EuroHPC JU infrastructure, supported by Mimer AI Factory, demonstrating that world-class AI innovation can and is being built in Europe.
“A tokenizer determines how well an AI model understands and processes human language,” says Sonny Mir, founder of TerraNex AB. “These results validate our architectural approach and give us a strong foundation for training the full TNex AI model.” The tokenizer has been tested across a wide range of European languages, laying the groundwork for an AI system that treats all languages and regions with equal importance, which is a core TerraNex mission, says Sonny.
Mimer is very happy to support this work since we believe that it is a significant step toward a trustworthy, multilingual AI platform built for European companies, public institutions, and citizens.
Learn more at terranex.ai.