German AI Consortium Launches Soofi S 30B-A3B, Outperforms on English and German Benchmarks

ME News reports that on July 13 (UTC+8), a research consortium coordinated by the German AI Association released the open-source large language model Soofi S 30B-A3B. The model has a total of 31.6 billion parameters, with only approximately 3.2 billion parameters activated per token, employing a MoE architecture combining Mamba-2 and standard attention layers. The model was fully trained on Deutsche Telekom’s Munich Industrial AI Cloud, with German language data increasing from 7.2% in the first phase to 15.3% in the second phase. On benchmark tests, Soofi S achieved the highest combined score in both English and German among all fully open-source models, outperforming OLMo 3 32B and Apertus 70B. It scored 73.8% on HumanEval, 70.2 on MBPP, and 84.2 on the German version of MBPP. The model supports a context window of up to 1 million tokens and achieves approximately eight times the generation throughput of comparable dense models at a 40,000-token length. The model weights are now open-sourced. 🔗 Read the original article: https://the-decoder.com/german-ai-consortium-releases-soofi-s-an-open-30b-model-that-tops-benchmarks-in-both-english-and-german via AI HOT · https://aihot.virxact.com/items/cmrj6actv0651bilkm5pfz6ub (Source: AiHot)