German AI consortium releases Soofi S, an open 30B model that tops benchmarks in both English and German

A German research consortium coordinated by the KI Bundesverband has released Soofi S, an open 30B language model that tops benchmarks in both English and German among fully open models, the consortium announced. The model, formally named Soofi S 30B-A3B, uses a hybrid mixture-of-experts architecture with 31.6 billion total parameters that activates only 3.2 billion per token. Training took place between March and May on up to 512 Nvidia B200 GPUs at Deutsche Telekom's Industrial AI Cloud in Munich, totaling about 253,000 GPU-hours. The consortium processed about 27 trillion tokens across three phases, with German data rising from 7.2 percent in the first phase to 15.3 percent in the second. In evaluations against 16 other open models, Soofi S leads fully open models on aggregate scores for German and English, surpassing OLMo 3 32B and Apertus 70B. It also leads on code benchmarks including HumanEval and MBPP. The model shows weaker performance on German competition math and factual retrieval. Participants include Fraunhofer Institutes IAIS and IIS, DFKI, TU Darmstadt, University of Wuerzburg, L3S Research Center, Berlin University of Applied Sciences, Ellamind, and Merantix Momentum. The project is funded by the German Federal Ministry for Economic Affairs and Energy under the European IPCEI-CIS program. Researchers are releasing model weights, training code, and a detailed data inventory.