--- title: "Soofi S: Is Germany's Sovereign LLM Ready?" canonical: https://wavect.io/blog/soofi-s-european-sovereign-llm/ language: en description: "Soofi S is Germany's 31.6B sovereign LLM, currently a selected closed beta. We examine benchmarks, license, hardware and deployment readiness." image: "https://wavect.io/img/blog/headers/header_soofi-s-european-sovereign-llm.png" --- [**Back**](/blog/overview/) [![Kevin Riedl](/img/team/kevin.webp)](/team/kevin-riedl/) [Kevin Riedl](/team/kevin-riedl/) https://linkedin.com/in/wsdt 14 min read · 16 Jul 2026 Last reviewed August 7, 2026 [**Next**](/blog/open-weight-llm-comparison-2026/) # Soofi S: Is Germany's Sovereign LLM Ready for Business? TL;DR Soofi S is a 31.6B-total, roughly 3.2B-active German-English Mixture-of-Experts foundation model trained from scratch on Deutsche Telekom's Industrial AI Cloud in Munich. Its current card documents a 1M-token context, about 253,000 B200 GPU-hours, and multiple Instruct, reasoning and quantized variants. The consortium's own report places it ahead of OLMo 3 32B and Apertus 70B among the fully open models tested, while Qwen3.5 scores higher in the broader comparison. As of 7 August 2026, however, the official checkpoint is explicitly a selected-partner closed beta, not an open release, and the promised permissive final license is still absent. Companies should seek a controlled partner pilot, freeze the artifact, run German and English task evals, and wait for the final ungated license before production procurement. Related service: [AI Enablement](/services/ai-enablement/) **Soofi S is a new German-English foundation model built by a German research consortium, trained from scratch on Deutsche Telekom's Industrial AI Cloud in Munich.** The current model card specifies 31.6 billion total parameters and roughly 3.2 billion active parameters per token, or about 3.6 billion including embeddings. It consumed about 26.68 trillion training tokens. The consortium's own evaluation puts it ahead of OLMo 3 32B and Apertus 70B on aggregate English and German scores among the fully open models it tested. That makes Soofi S technically important. It does not yet make it a production-ready European replacement for Claude, GPT, or every Chinese open-weight model. As of **7 August 2026**, the official base-model card explicitly calls the checkpoint a **closed beta for already selected partners**, not an open release. Access still requires sharing contact information, and the license field says only that a permissive license will accompany the final release. The commercial answer is therefore more useful than the launch headline: **Soofi S belongs on the pilot shortlist for German-language industrial AI, but not yet on an unqualified production shortlist.** This review separates what the project has demonstrated from what a buyer still needs to verify. | Question | Current answer | Commercial meaning | | --- | --- | --- | | Is it European? | Developed by a German consortium and trained in Munich | Strong European development and infrastructure provenance | | Is it the first European LLM? | No | Teuken, EuroLLM, Apertus, and other European projects came earlier | | Is it fully open? | No, the current release is a selected-partner closed beta | Wait for the final license and released training artifacts before approving commercial use | | Is it the best open model? | No universal ranking supports that claim | It leads a fully open subset in the authors' aggregate evaluation, while Qwen3.5 leads the broader comparison | | Can companies test it? | Only selected closed-beta partners can access the current checkpoint | Ask the consortium for pilot access; do not assume a public download path | ## What is Soofi S? [Soofi](https://www.soofi.info/) stands for Sovereign Open Source Foundation Models. The project is coordinated by the German AI Association and brings together Fraunhofer IAIS and IIS, DFKI, TU Darmstadt, the University of Würzburg, Leibniz University Hannover, Berlin University of Applied Sciences, ellamind, and Merantix Momentum. It is funded by Germany's Federal Ministry for Economic Affairs and Energy. The first model, Soofi S 31.6B-A3.2B, is a sparse Mixture-of-Experts model. It combines 23 Mamba-2 sequence-mixing layers, 23 sparse MoE layers, and six grouped-query attention layers. Each MoE layer routes to six of 128 experts plus two shared experts. Active parameters drive much of the compute per generated token, while the 31.6 billion total parameters still determine weight memory and storage. The architecture is aimed at long documents and concurrent inference. Only a small number of layers maintain a conventional attention cache, so the cache grows much more slowly than it does in a dense Transformer. In the [Soofi S pretraining report](https://arxiv.org/abs/2607.09424), the authors measured up to 9.2 times the aggregate decode throughput of dense 14B to 24B models at 40,000 tokens and batch size 32 on a single NVIDIA B200. That is a specific serving test, not a promise that every deployment will be nine times faster. The training provenance is equally central to the project. Soofi S was trained end to end on up to 512 B200 GPUs in Deutsche Telekom's Munich facility. [Deutsche Telekom describes the Industrial AI Cloud](https://www.telekom.com/en/media/media-information/archive/germany-s-first-ai-factory-for-industry-1101670) as infrastructure for European industry, research, start-ups, and the public sector. For a buyer, that proves where the pretraining happened. It does not automatically prove where your prompts, retrieval data, logs, or backups will be processed after deployment. ## Is Soofi S really Europe's first fully European LLM? **No. "Europe's first LLM" is not an accurate description.** Europe already had several models developed and trained by European teams. The German OpenGPT-X project published [Teuken-7B](https://arxiv.org/abs/2410.03730), covering all 24 official EU languages, in 2024. [EuroLLM-22B](https://arxiv.org/abs/2602.05879) was trained from scratch for 24 EU languages and 11 additional languages. Switzerland's [Apertus 70B](https://huggingface.co/swiss-ai/Apertus-70B-2509) released weights, data reconstruction code, intermediate checkpoints, and training recipes under Apache 2.0. The defensible wording is narrower: **Soofi S is the first model in the Soofi family, a German-developed and German-trained sovereign foundation model with unusually ambitious openness and strong German-English performance.** That is still significant. It just is not a historical first for European language models. It is also not "fully European" if that phrase means every dependency came from Europe. The model was trained in Germany and governed by a German consortium, but the training stack and hardware include NVIDIA technology. Sovereignty is not autarky. The practical goal is control over the model artifacts, infrastructure, data flows, and right to operate without one non-European model vendor deciding access or price. ## Is Soofi S fully open-source or only open-weight? The project's openness target goes well beyond weights. The pretraining report says the final release will include model weights, selected intermediate checkpoints, per-source data accounting, hyperparameters, and training and evaluation code. Where source licenses permit, the consortium plans to release data-construction artifacts. Commercially licensed sources are documented through mixture accounting rather than redistributed as raw text. That direction aligns more closely with the [Open Source AI Definition](https://opensource.org/ai/open-source-ai-definition) than a normal open-weight release does. The OSI definition expects the parameters, the code used to train and run the system, and sufficiently detailed information about the training data to let a skilled team recreate an equivalent system. But buyers procure what exists, not what a paper promises to publish. The [current official base-model card](https://huggingface.co/Soofi-Project/Soofi-S-Base) now provides more engineering detail than it did at launch, while stating: - the checkpoint is a closed-beta research artifact for already selected partners; - the model is an unaligned base model, not a safe end-user assistant; - the final model will be released under a permissive license, but that license is not yet attached; - downloads remain gated behind contact-information sharing; - the card now documents a 1,048,576-token context, about 253,000 B200 GPU-hours, the March-to-May training run, and multiple post-trained and quantized preview variants; - energy/emissions, prescribed quality standards, and some evaluation fields still remain TODOs. So the precise answer on 7 August 2026 is: **Soofi S is designed to become a fully open model, but the current closed beta is not an open commercial release.** Legal and security teams should wait for the final license, ungated artifact inventory, complete model card, and immutable release hashes. ## How strong is Soofi S on German, English, coding, and reasoning? The benchmark headline is real within a defined category. In the consortium's evaluation of 17 open base models, Soofi S scored highest among the models classified as fully open on the authors' English and German aggregates. It ranked ahead of OLMo 3 32B and Apertus 70B, and it led the tested fully open group on code aggregates in both languages. | Benchmark | Soofi S | What it indicates | Caveat | | --- | --- | --- | --- | | English aggregate | 70.1 | Competitive with similarly sized open-weight models | Qwen3.5 35B-A3B scored 74.6 | | German aggregate | 79.1 | Strong German performance | Qwen3.5 35B-A3B scored 81.6 | | HumanEval | 73.8 | Strong isolated code generation | Does not measure production repository work | | MBPP | 70.2 | Strong short programming tasks | Small synthetic tasks differ from real maintenance | | MBPP-DE | 84.2 | Best result in the authors' comparison | Self-reported and not yet independently reproduced | This is where the LinkedIn claim needs surgery. Soofi S does not top a universal ranking of all open models. It tops the fully open subset used in the Soofi team's aggregate evaluation. In the broader open-weight comparison, [Qwen3.5 35B-A3B](https://huggingface.co/Qwen/Qwen3.5-35B-A3B) scored higher on both the English and German aggregates. Soofi S also trails Qwen3.5 on German competition mathematics, while performing particularly well on several coding measures and German regional knowledge. The authors also report a long-context weakness. The architecture keeps serving throughput high as context grows, but a RULER extraction task involving frequent words deteriorated sharply beyond 32,000 tokens. Fast long-context generation and accurate long-context retrieval are different properties. A procurement test needs to measure both. None of these base-model results proves that Soofi S matches the best Claude, GPT, Gemini, GLM, or Qwen production assistant on an end-to-end business workflow. Base-model benchmarks isolate pretraining quality. A production assistant also depends on post-training, tool use, retrieval, guardrails, serving reliability, and support. ## Where could Soofi S create commercial value? The model is most interesting when German-language quality, infrastructure control, and transparency matter more than winning every frontier benchmark. - **Technical and regulatory document workflows.** German manuals, standards, policies, tenders, and internal knowledge bases are a natural evaluation target. Use retrieval and citation checks rather than expecting the model to contain current documents. - **Industrial coding assistants.** The reported code scores justify a pilot for code explanation, test generation, migration support, and internal developer tooling. Repository-level acceptance tests still decide whether it ships. - **Agentic back-office workflows.** The sparse architecture may suit concurrent agents processing long documents, provided tool-use accuracy, permissions, and failure handling pass a real eval. - **Domain adaptation.** A transparent base model can be continued-pretrained or fine-tuned for an industry's language without depending on a proprietary vendor's roadmap. - **Public-sector and regulated pilots.** European training provenance and the option to operate on controlled infrastructure can simplify part of the sovereignty discussion. It does not remove procurement, security, data protection, or model-risk work. For these use cases, model choice is only one layer. Wavect's [AI enablement work](/services/ai-enablement/) covers the surrounding system: use-case selection, data access, retrieval, evaluation, routing, deployment, and knowledge transfer. The proof pattern is visible in our [Twinsoft AI case study](/case-studies/twinsoft-ai/), where the valuable part was the production system around the model, not a leaderboard position. If the sovereignty argument is what makes this model interesting to you, the deployment question follows immediately after the model question. That is the practical side of [AI software development in Austria](/ai-software-development-austria/): mapping the data flow first, then choosing EU-region or self-hosted deployment where the stack actually supports it, rather than assuming a European model automatically settles the compliance discussion. ## When should a company not choose Soofi S? Do not choose it today when any of these conditions is true: - **You need a signed commercial license or immediate public access.** The closed-beta model card does not yet provide either. - **You need a vendor SLA and managed support.** The current public release is a research preview, not a managed enterprise service. - **Your workload is mainly multimodal or highly multilingual.** Soofi S focuses on German and English. Qwen3.5, Apertus, or EuroLLM may fit broader language or modality requirements better. - **You need the strongest available abstract reasoning.** The model's strongest evidence is German, coding, and serving efficiency, not universal frontier leadership. - **Your volume is low and your data can use a managed API.** Running a 30B-total-parameter model can cost more in engineering and idle infrastructure than the API bill it replaces. Use our [local-model versus API break-even calculator](/blog/local-models-vs-apis-break-even-eu-2026/) for that decision. - **You cannot build and maintain an eval harness.** A sovereign model without measured task quality is still an unmanaged production risk. ## What should buyers verify before a Soofi S pilot? 1. **Freeze the exact artifact.** Record the repository, revision, weight hashes, model card, custom code, tokenizer, and quantization. Preview artifacts can change. 2. **Clear the final license.** Confirm commercial use, modification, redistribution, output terms, and obligations for derivatives. Do not infer rights from the project name. 3. **Choose the right variant.** The base model is not a chatbot. Test the Instruct preview for general assistant work and evaluate reasoning variants separately rather than transferring base-model scores to them. 4. **Build a use-case eval.** Include at least 50 to 100 representative German and English tasks, hard negatives, unacceptable outputs, citation checks, tool-call accuracy, and human scoring. 5. **Benchmark the target serving stack.** Measure prefill, time to first token, decode throughput, concurrency, total GPU memory, failure recovery, and quantization loss on the hardware you will actually operate. 6. **Map the entire data path.** Document embeddings, vector storage, prompt logs, observability, backups, support access, and fallback APIs. A European model can still sit inside a non-sovereign system. 7. **Price the successful task.** Compare Soofi S with at least one European fully open alternative, one permissively licensed open-weight model, and one managed API. Include engineering, redundancy, monitoring, and eval maintenance. Our [technology selection guide](/software-development-guide/how-to-choose-a-tech-stack-for-mvp/) explains the broader principle: choose against the constraint that is expensive to reverse. For Soofi S, that is currently license and operational maturity first, task quality second, then throughput and cost. ## Soofi S vs Apertus, OLMo 3, and Qwen3.5 | Model | Origin | Openness | Best reason to test | Main trade-off | | --- | --- | --- | --- | --- | | Soofi S 31.6B-A3.2B | Germany | Full openness intended; current checkpoint is a selected-partner closed beta | German, code, sparse long-context serving, European provenance | No open commercial release or final license yet | | Apertus 70B | Switzerland | Apache 2.0 with open data and training resources | Mature fully open European option with broad language coverage | Dense 70B deployment is much heavier | | OLMo 3 32B | United States | Apache 2.0 with code, data, checkpoints, and recipes | Open-science baseline with mature reproducibility | English-first and not a European sovereignty story | | Qwen3.5 35B-A3B | China | Apache 2.0 weights, not a fully open training stack | Stronger broader aggregate scores, multimodality, mature ecosystem | Non-European model governance and less training transparency | This table does not replace our broader [open-weight LLM comparison](/blog/open-weight-llm-comparison-2026/). That page owns the market-wide choice among DeepSeek, Qwen, Kimi, GLM, and Llama. This page owns one narrower decision: whether Soofi S is ready for a European organization's pilot and procurement process. ## Frequently Asked Questions About Soofi S ### Is Soofi S a German LLM or a European LLM? Both descriptions can be accurate with context. It was developed by a German consortium, trained on infrastructure in Munich, and funded through German and European programmes. Its positioning is European because it is intended as sovereign infrastructure for European industry and administration. ### Can Soofi S be used commercially today? Do not assume so from the project name. The current base-model card calls the checkpoint a closed beta for selected partners and says the final model will later use a permissive license. A commercial team should obtain explicit pilot terms now and wait for the final license before production use. ### Can Soofi S run locally? Preview checkpoints include BF16, FP8, and community-facing GGUF paths for tools such as vLLM, llama.cpp, and Ollama. Sparse activation reduces compute per token, but the machine must still store and load the total model weights. Benchmark your chosen quantization on your own hardware. ### Is Soofi S better than Qwen3.5? Not overall. Soofi S is strong on German and several code benchmarks, and it leads the fully open group in the consortium's aggregate evaluation. Qwen3.5 35B-A3B scores higher on the report's broader English and German aggregates and offers a more mature post-trained, multimodal ecosystem. ### Does using Soofi S make an AI system sovereign? No. The model removes one dependency. Sovereignty also depends on where inference runs, who controls the infrastructure and encryption keys, where retrieval data and logs live, which fallback APIs receive prompts, and whether your team can operate or replace every critical layer. ## Final thoughts Soofi S is not Europe's first LLM, not the world's best open model, and not yet a procurement-ready fully open commercial release. It is a credible German-English foundation model trained on German infrastructure, with strong code and German benchmarks, an efficient sparse architecture, a documented 1M-token context, and a release plan that aims to expose far more than weights. For European companies, the right move is to seek a controlled partner pilot, not treat the closed beta as a public release. Freeze the checkpoint, wait for the final license, evaluate real German and English tasks, benchmark the exact serving stack, and compare cost per successful task with Apertus, Qwen3.5, and a managed frontier API. If Soofi S wins that test, its sovereignty story becomes a business advantage. Until then, it is a promising candidate, not a default. ## You may also like.. [**Open-Weight LLM Showdown 2026** Compare the wider market on license, coding, reasoning, context, price, and EU self-host viability.](/blog/open-weight-llm-comparison-2026/) [**AI Enablement vs a generic AI consultancy** A model shortlist is not a production system. Compare recommendations with implementation, evals, and handover.](/compare/ai-enablement-vs-generic-ai-consultancy/) Models and infrastructure ## Continue through this cluster Model selection, inference economics, local deployment, compression and serving architecture. [Start with the cornerstone**Self-Hosting LLMs in the EU: When Open Weights Actually Pay Off**](/blog/self-hosting-llms-eu-cost/) - [Ox Alpha Free AI Model: Setup, Privacy and Buyer Guide](/blog/ox-alpha-free-ai-model-guide-2026/) - [Pika Audio Models API Pricing: Is SFX Really 20x Cheaper?](/blog/pika-audio-models-api-pricing-2026/) - [Thunder Compute's $13M GPU Virtualization Bet: Enterprise Buyer's Guide](/blog/thunder-compute-gpu-virtualization-series-a/) - [AirLLM on 4 GB VRAM: How Layer-Wise Inference Really Works](/blog/airllm-layer-wise-inference-low-vram/) - [Qwen3.8-27B: Self-Hosted Computer-Use Agents Without Exporting Screenshots](/blog/qwen3-8-27b-self-hosted-computer-use-agents/) Inbox, without the noise ## Follow the work that matters to you Get a short email when we publish something new. Follow the whole blog or only the problems you care about. [**Back**](/blog/overview/) [![Kevin Riedl](/img/team/kevin.webp)](/team/kevin-riedl/) [Kevin Riedl](/team/kevin-riedl/) https://linkedin.com/in/wsdt 14 min read · 16 Jul 2026 Last reviewed August 7, 2026 [**Next**](/blog/open-weight-llm-comparison-2026/) New posts by email × × Get new posts by email A short email when we publish. Free, no tracking. ## Structured Data ```json { "@context": "https://schema.org", "@graph": [ { "@id": "https://wavect.io/#organization", "@type": [ "Organization", "ProfessionalService", "LocalBusiness" ], "employee": [ { "@id": "https://wavect.io/team/kevin-riedl/#person", "@type": "Person", "jobTitle": "Managing Director", "name": "Kevin Riedl", "url": "https://wavect.io/team/kevin-riedl/", "worksFor": { "@id": "https://wavect.io/#organization", "@type": [ "Organization", "ProfessionalService", "LocalBusiness" ] } }, { "@id": "https://wavect.io/team/christof-jori/#person", "@type": "Person", "jobTitle": "Managing Director", "name": "Christof Jori", "url": "https://wavect.io/team/christof-jori/", "worksFor": { "@id": "https://wavect.io/#organization", "@type": [ "Organization", "ProfessionalService", "LocalBusiness" ] } } ], "founder": [ { "@id": "https://wavect.io/team/kevin-riedl/#person", "@type": "Person", "jobTitle": "Managing Director", "name": "Kevin Riedl", "url": "https://wavect.io/team/kevin-riedl/", "worksFor": { "@id": "https://wavect.io/#organization", "@type": [ "Organization", "ProfessionalService", "LocalBusiness" ] } }, { "@id": "https://wavect.io/team/christof-jori/#person", "@type": "Person", "jobTitle": "Managing Director", "name": "Christof Jori", "url": "https://wavect.io/team/christof-jori/", "worksFor": { "@id": "https://wavect.io/#organization", "@type": [ "Organization", "ProfessionalService", "LocalBusiness" ] } } ], "legalRepresentative": [ { "@id": "https://wavect.io/team/kevin-riedl/#person", "@type": "Person", "jobTitle": "Managing Director", "name": "Kevin Riedl", "url": "https://wavect.io/team/kevin-riedl/", "worksFor": { "@id": "https://wavect.io/#organization", "@type": [ "Organization", "ProfessionalService", "LocalBusiness" ] } }, { "@id": "https://wavect.io/team/christof-jori/#person", "@type": "Person", "jobTitle": "Managing Director", "name": "Christof Jori", "url": "https://wavect.io/team/christof-jori/", "worksFor": { "@id": "https://wavect.io/#organization", "@type": [ "Organization", "ProfessionalService", "LocalBusiness" ] } } ], "name": "Wavect GmbH", "subjectOf": { "@id": "https://wavect.io/verified-claims.json#dataset", "@type": "Dataset", "creator": { "@id": "https://wavect.io/#organization", "@type": [ "Organization", "ProfessionalService", "LocalBusiness" ] }, "description": "A machine-readable registry of quantitative and qualitative claims published by Wavect, with review dates, localized page appearances and public third-party citations where available.", "inLanguage": "en", "isAccessibleForFree": true, "license": "https://creativecommons.org/licenses/by/4.0/", "name": "Wavect verified publication claims", "url": "https://wavect.io/verified-claims.json" }, "url": "https://wavect.io/" }, { "@id": "https://wavect.io/team/kevin-riedl/#person", "@type": "Person", "jobTitle": "Managing Director", "name": "Kevin Riedl", "sameAs": [ "https://www.wikidata.org/wiki/Q139796365", "https://www.linkedin.com/in/wsdt", "https://github.com/wsdt" ], "url": "https://wavect.io/team/kevin-riedl/", "worksFor": { "@id": "https://wavect.io/#organization", "@type": [ "Organization", "ProfessionalService", "LocalBusiness" ] } }, { "@id": "https://wavect.io/team/christof-jori/#person", "@type": "Person", "jobTitle": "Managing Director", "name": "Christof Jori", "sameAs": [ "https://www.wikidata.org/wiki/Q139796367", "https://www.linkedin.com/in/jocr77/", "https://github.com/jo-chris" ], "url": "https://wavect.io/team/christof-jori/", "worksFor": { "@id": "https://wavect.io/#organization", "@type": [ "Organization", "ProfessionalService", "LocalBusiness" ] } }, { "@id": "https://wavect.io/#website", "@type": "WebSite", "inLanguage": [ "en", "de", "es", "zh" ], "name": "Wavect", "potentialAction": { "@type": "SearchAction", "query-input": "required name=search_term_string", "target": { "@type": "EntryPoint", "urlTemplate": "https://wavect.io/search/?q={search_term_string}" } }, "publisher": { "@id": "https://wavect.io/#organization", "@type": [ "Organization", "ProfessionalService", "LocalBusiness" ] }, "url": "https://wavect.io/" }, { "@id": "https://wavect.io/blog/soofi-s-european-sovereign-llm/#webpage", "@type": "WebPage", "dateModified": "2026-07-16", "inLanguage": "en", "isPartOf": { "@id": "https://wavect.io/#website", "@type": "WebSite" }, "lastReviewed": "2026-07-16", "url": "https://wavect.io/blog/soofi-s-european-sovereign-llm/" } ] } ``` ```json { "@context": "https://schema.org", "@type": "BlogPosting", "abstract": "Soofi S is a 31.6B-total, roughly 3.2B-active German-English Mixture-of-Experts foundation model trained from scratch on Deutsche Telekom's Industrial AI Cloud in Munich. Its current card documents a 1M-token context, about 253,000 B200 GPU-hours, and multiple Instruct, reasoning and quantized variants. The consortium's own report places it ahead of OLMo 3 32B and Apertus 70B among the fully open models tested, while Qwen3.5 scores higher in the broader comparison. As of 7 August 2026, however, the official checkpoint is explicitly a selected-partner closed beta, not an open release, and the promised permissive final license is still absent. Companies should seek a controlled partner pilot, freeze the artifact, run German and English task evals, and wait for the final ungated license before production procurement.", "articleBody": " Blog overview/AI and agents/Models and infrastructure Soofi S: Is Germany's Sovereign LLM Ready for Business? TL;DR Soofi S is a 31.6B-total, roughly 3.2B-active German-English Mixture-of-Experts foundation model trained from scratch on Deutsche Telekom's Industrial AI Cloud in Munich. Its current card documents a 1M-token context, about 253,000 B200 GPU-hours, and multiple Instruct, reasoning and quantized variants. The consortium's own report places it ahead of OLMo 3 32B and Apertus 70B among the fully open models tested, while Qwen3.5 scores higher in the broader comparison. As of 7 August 2026, however, the official checkpoint is explicitly a selected-partner closed beta, not an open release, and the promised permissive final license is still absent. Companies should seek a controlled partner pilot, freeze the artifact, run German and English task evals, and wait for the final ungated license before production procurement. Related service: AI Enablement Soofi S is a new German-English foundation model built by a German research consortium, trained from scratch on Deutsche Telekom's Industrial AI Cloud in Munich. The current model card specifies 31.6 billion total parameters and roughly 3.2 billion active parameters per token, or about 3.6 billion including embeddings. It consumed about 26.68 trillion training tokens. The consortium's own evaluation puts it ahead of OLMo 3 32B and Apertus 70B on aggregate English and German scores among the fully open models it tested. That makes Soofi S technically important. It does not yet make it a production-ready European replacement for Claude, GPT, or every Chinese open-weight model. As of 7 August 2026, the official base-model card explicitly calls the checkpoint a closed beta for already selected partners, not an open release. Access still requires sharing contact information, and the license field says only that a permissive license will accompany the final release. The commercial answer is therefore more useful than the launch headline: Soofi S belongs on the pilot shortlist for German-language industrial AI, but not yet on an unqualified production shortlist. This review separates what the project has demonstrated from what a buyer still needs to verify. Soofi S in one minute, verified 7 August 2026 QuestionCurrent answerCommercial meaning Is it European?Developed by a German consortium and trained in MunichStrong European development and infrastructure provenance Is it the first European LLM?NoTeuken, EuroLLM, Apertus, and other European projects came earlier Is it fully open?No, the current release is a selected-partner closed betaWait for the final license and released training artifacts before approving commercial use Is it the best open model?No universal ranking supports that claimIt leads a fully open subset in the authors' aggregate evaluation, while Qwen3.5 leads the broader comparison Can companies test it?Only selected closed-beta partners can access the current checkpointAsk the consortium for pilot access; do not assume a public download path What is Soofi S? Soofi stands for Sovereign Open Source Foundation Models. The project is coordinated by the German AI Association and brings together Fraunhofer IAIS and IIS, DFKI, TU Darmstadt, the University of Würzburg, Leibniz University Hannover, Berlin University of Applied Sciences, ellamind, and Merantix Momentum. It is funded by Germany's Federal Ministry for Economic Affairs and Energy. The first model, Soofi S 31.6B-A3.2B, is a sparse Mixture-of-Experts model. It combines 23 Mamba-2 sequence-mixing layers, 23 sparse MoE layers, and six grouped-query attention layers. Each MoE layer routes to six of 128 experts plus two shared experts. Active parameters drive much of the compute per generated token, while the 31.6 billion total parameters still determine weight memory and storage. The architecture is aimed at long documents and concurrent inference. Only a small number of layers maintain a conventional attention cache, so the cache grows much more slowly than it does in a dense Transformer. In the Soofi S pretraining report, the authors measured up to 9.2 times the aggregate decode throughput of dense 14B to 24B models at 40,000 tokens and batch size 32 on a single NVIDIA B200. That is a specific serving test, not a promise that every deployment will be nine times faster. The training provenance is equally central to the project. Soofi S was trained end to end on up to 512 B200 GPUs in Deutsche Telekom's Munich facility. Deutsche Telekom describes the Industrial AI Cloud as infrastructure for European industry, research, start-ups, and the public sector. For a buyer, that proves where the pretraining happened. It does not automatically prove where your prompts, retrieval data, logs, or backups will be processed after deployment. Is Soofi S really Europe's first fully European LLM? No. \"Europe's first LLM\" is not an accurate description. Europe already had several models developed and trained by", "articleSection": "Engineering", "author": { "@id": "https://wavect.io/team/kevin-riedl/#person", "@type": "Person", "name": "Kevin Riedl", "sameAs": [ "https://www.wikidata.org/wiki/Q139796365", "https://www.linkedin.com/in/wsdt", "https://github.com/wsdt" ], "url": "https://wavect.io/team/kevin-riedl/" }, "dateModified": "2026-08-07", "datePublished": "2026-07-16", "description": "Soofi S is a 31.6B-total, roughly 3.2B-active German-English Mixture-of-Experts foundation model trained from scratch on Deutsche Telekom's Industrial AI Cloud in Munich. Its current card documents a 1M-token context, about 253,000 B200 GPU-hours, and multiple Instruct, reasoning and quantized variants. The consortium's own report places it ahead of OLMo 3 32B and Apertus 70B among the fully open models tested, while Qwen3.5 scores higher in the broader comparison. As of 7 August 2026, however, the official checkpoint is explicitly a selected-partner closed beta, not an open release, and the promised permissive final license is still absent. Companies should seek a controlled partner pilot, freeze the artifact, run German and English task evals, and wait for the final ungated license before production procurement.", "headline": "Soofi S: Is Germany's Sovereign LLM Ready for Business?", "image": "https://wavect.io/img/blog/headers/header_soofi-s-european-sovereign-llm.svg", "inLanguage": "en", "mainEntityOfPage": { "@id": "https://wavect.io/blog/soofi-s-european-sovereign-llm/", "@type": "WebPage" }, "publisher": { "@id": "https://wavect.io/#organization", "@type": [ "Organization", "ProfessionalService", "LocalBusiness" ] }, "url": "https://wavect.io/blog/soofi-s-european-sovereign-llm/", "wordCount": 3001 } ``` ```json { "@context": "https://schema.org", "@type": "BreadcrumbList", "itemListElement": [ { "@type": "ListItem", "item": "https://wavect.io/", "name": "Home", "position": 1 }, { "@type": "ListItem", "item": "https://wavect.io/blog/overview/", "name": "Blog overview", "position": 2 }, { "@type": "ListItem", "item": "https://wavect.io/blog/topics/ai-agents/", "name": "AI and agents", "position": 3 }, { "@type": "ListItem", "item": "https://wavect.io/blog/clusters/models-infrastructure/", "name": "Models and infrastructure", "position": 4 }, { "@type": "ListItem", "item": "https://wavect.io/blog/soofi-s-european-sovereign-llm/", "name": "Soofi S: Is Germany's Sovereign LLM Ready? | ", "position": 5 } ] } ``` ```json { "@context": "https://schema.org", "@type": "FAQPage", "mainEntity": [ { "@type": "Question", "acceptedAnswer": { "@type": "Answer", "text": "Both descriptions can be accurate with context. It was developed by a German consortium, trained on infrastructure in Munich, and funded through German and European programmes. Its positioning is European because it is intended as sovereign infrastructure for European industry and administration." }, "name": "Is Soofi S a German LLM or a European LLM?" }, { "@type": "Question", "acceptedAnswer": { "@type": "Answer", "text": "Do not assume so from the project name. The current base-model card calls the checkpoint a closed beta for selected partners and says the final model will later use a permissive license. A commercial team should obtain explicit pilot terms now and wait for the final license before production use." }, "name": "Can Soofi S be used commercially today?" }, { "@type": "Question", "acceptedAnswer": { "@type": "Answer", "text": "Preview checkpoints include BF16, FP8, and community-facing GGUF paths for tools such as vLLM, llama.cpp, and Ollama. Sparse activation reduces compute per token, but the machine must still store and load the total model weights. Benchmark your chosen quantization on your own hardware." }, "name": "Can Soofi S run locally?" }, { "@type": "Question", "acceptedAnswer": { "@type": "Answer", "text": "Not overall. Soofi S is strong on German and several code benchmarks, and it leads the fully open group in the consortium's aggregate evaluation. Qwen3.5 35B-A3B scores higher on the report's broader English and German aggregates and offers a more mature post-trained, multimodal ecosystem." }, "name": "Is Soofi S better than Qwen3.5?" }, { "@type": "Question", "acceptedAnswer": { "@type": "Answer", "text": "No. The model removes one dependency. Sovereignty also depends on where inference runs, who controls the infrastructure and encryption keys, where retrieval data and logs live, which fallback APIs receive prompts, and whether your team can operate or replace every critical layer." }, "name": "Does using Soofi S make an AI system sovereign?" } ] } ```