German AI consortium releases Soofi S, an open 30B model that tops benchmarks
the-decoder.com
July 16, 2026
7 min read
🔥🔥🔥🔥🔥
52/100
Summary
The German research consortium has released Soofi S, an open-source language model with 30 billion parameters, trained on Deutsche Telekom's AI cloud infrastructure. The model features a hybrid architecture that activates only 3.2 of its 31.6 billion parameters per token, maintaining processing speed with long inputs and excelling in both English and German benchmarks.
Key Takeaways
- The German AI consortium released Soofi S, an open-source language model with 30 billion parameters, trained entirely on Deutsche Telekom's AI cloud infrastructure.
- Soofi S utilizes a hybrid architecture that activates only 3.2 billion parameters per token, maintaining processing speed for long inputs.
- The model outperforms other fully open models in benchmarks for English and German, surpassing leaders like OLMo 3 32B and Apertus 70B.
- Soofi S was trained on approximately 27 trillion tokens, with a training mix heavily weighted towards German data.
Community Sentiment
Positives
- Great to see more competition in this space, and especially from Europe, but it’s disappointing the benchmarks don’t include the current best comparable models.
- It's refreshing to see model training powered entirely by renewable energy — a step towards a sustainable AI future!
Concerns
- It tops benchmarks because it uses them in its training data — that's a major red flag for credibility.
- Disappointing. Sounds like Volkswagen 2.0 — concerns about ethical practices are looming large.
- The benchmarks are outdated, showing results against models that have been surpassed — feels like a missed opportunity.
