Deutsche Telekom has released Soofi S 30B-A3B, a sovereign, open-source Mixture-of-Experts hybrid Mamba Transformer foundation model designed for German and English. Built end-to-end on the German Industrial AI Cloud, the model activates only 3 billion of its 30 billion parameters per token to maintain near-constant inference cache as context grows.
- Pretrained on roughly 27 trillion tokens with up-weighted German data, Soofi S matches dense 14 to 27B models on aggregate English and German benchmarks.
- It achieves the best code aggregates in both languages among 17 open base models and outperforms every European sovereign baseline in comparisons.
- Among fully open models, it obtains the highest English and German evaluation scores, ahead of Olmo 3 32B and Apertus 70B.
- The release includes weights, selected intermediate checkpoints, full per-source data accounting, hyperparameters, and training and evaluation code under highly permissive terms.
The model offers a decisive throughput advantage over dense models for long-context, high-concurrency deployment while providing comprehensive transparency through open-access data construction artifacts.