On July 13, 2026, a German research consortium released Soofi S, an open 30B-parameter language model that now leads every fully open alternative on both German and English benchmarks. It is the first release from the Soofi project, a consortium coordinated by the KI Bundesverband with Fraunhofer institutes, DFKI, universities, and companies including Ellamind and Merantix Momentum. The German Federal Ministry for Economic Affairs and Energy funded the project with roughly 20 million euros. This post looks at what actually shipped, where the model is strong, where it is not, and what it takes to run it behind a production agent today.
What shipped
The release covers the base model (Soofi-S-30B-A3B on Hugging Face), an instruction-tuned preview, two reasoning previews named Isar and Rhine, intermediate training checkpoints, and the full training and evaluation code. Training ran from late March to mid May 2026 on up to 512 Nvidia B200 GPUs in Munich, about 253,000 GPU-hours on Deutsche Telekom infrastructure, over roughly 27 trillion tokens. German content was deliberately overweighted: 7.2% of the main pretraining mix and 15.3% of the later high-quality annealing phase, compared with the roughly 5% that common training recipes reserve for all non-English data combined.
The architecture is the interesting part
Soofi S is not chasing frontier scores. It is designed to be cheap to serve. The model is a hybrid: of its 52 layers, only 6 are classic attention layers that keep a KV cache, while the rest pair Mamba-2 state-space layers with mixture-of-experts blocks. The practical consequence is that throughput barely degrades as the context grows. The consortium measured nearly constant decode speed from 4,000 out to 256,000 tokens of context: around 4,800 tokens per second per GPU at a 40,000-token context with 32 parallel requests, several times what dense models in the 14-24B range manage under the same load.
For agent workloads this is the headline feature. Agents accumulate context: system prompts, tool schemas, tool results, retrieved documents, conversation history. A model whose serving cost stays flat as that context grows is built for exactly this shape of traffic. The consortium says as much: the stated focus for the next project phase is technical documents, code generation, and agent-based systems.
The benchmarks, honestly
Among fully open models, meaning weights, data recipe, and training code are all public, Soofi S is now the strongest on both German and English aggregates, ahead of OLMo 3 32B and Apertus 70B. Its German results are the standout: an aggregate of 79.1 and the best German code scores published for an open model. The technical report is unusually transparent about the rest.
Selected scores from the Soofi S technical report, pass@1 for code benchmarks.
The caveats matter just as much. Against open-weight but not fully open models, Soofi S loses clearly: Qwen3.5 scores 74.6 on the English aggregate to Soofi's 70.1, with similar gaps on abstract reasoning benchmarks like GPQA. Competition-style math in German is weak at 56 points on Minerva MATH-DE, and open-domain factual recall runs into the limits of ~3.2B active parameters. And while long-context throughput is excellent, one RULER word-extraction subtask degrades past 32,000 tokens and drops below 5% accuracy at the longest lengths, so the 1M-token context is best treated as a throughput property, not a guarantee of reliable recall across it.
Critics have also noted that the architecture overlaps heavily with Nvidia's Nemotron 3 Nano family and argue that continual pretraining on an existing base would have cost far less than the 253,000 GPU-hours spent. That is a fair critique of the economics. It does not change what the released artifact is: a fully reproducible, German-strong open model that anyone can inspect, retrain, or build on.
What open means here
Soofi S meets the Open Source AI Definition 1.0 from the Open Source Initiative. About 99% of the training run can be independently reconstructed from the published data accounting; the remainder is commercially licensed German newspaper archive content that cannot be redistributed. Weights, intermediate checkpoints, training code, and evaluation code are all public, which puts it in a small club alongside OLMo and Apertus rather than the much larger group of weights-only releases.
Running Soofi S behind an agent
Because Connic is bring-your-own-key and never hosts models itself, an open model you serve yourself plugs in the same way any provider does. Serve Soofi S with vLLM or, once the quantized variants land, llama.cpp or Ollama, register the endpoint as a custom OpenAI-compatible provider in your project settings, and reference it from the agent config:
model: vllm/soofi-s-instruct. If the inference box lives inside your private network, agents reach it through the bridge instead of the public internet. Guardrails, judges, traces, and approvals work the same regardless of which model answers.The honest caveats carry through. The base model is explicitly not meant for direct use without post-training, the instruct and reasoning variants are previews, and there is no commercial support behind the project: patches, security updates, and incident response are yours. That is the standard trade of self-hosting, and it is why we would put Soofi S on a slice of real traffic behind an A/B test, with judges scoring its output, before routing anything important to it. Our DACH report found 23% of production agents in the region already run open-weight or self-hosted models, so the operational pattern is well established; what has been missing is a German-first model worth the effort.
Connect a self-hosted Soofi S endpoint as a custom provider and put guardrails, judges, and traces around it, alongside whichever commercial models you already use.
Get started freeThe sovereignty question
Soofi S will be discussed mostly in sovereignty terms, and it is worth being precise about what it does and does not settle. It does not make the strongest closed or open-weight models redundant; the benchmark gaps are real. What it does provide is something that did not exist before: a competitive, fully reproducible model that is strong in German, trained in the EU, and able to run entirely on infrastructure you control. For teams whose deployments hinge on EU AI Act readiness and keeping data inside the EU, that combination is the point, independent of leaderboard positions.
Three things to watch from here: the final license text, the instruct model graduating from preview, and the consortium's next phase, which targets agent systems directly and is looking for industry partners. If those land, the default answer to "which model do we self-host for German workloads" is about to change.