Soofi S is an open ~30B German consortium LLM built for sovereign AI. Honest comparison vs Kompozy: when a self-hosted model wins, and when you need a content engine.
If you searched "Soofi S vs Kompozy," the first honest thing to say is that they are not the same class of product. Soofi S is a foundation model — a set of open weights you download and run. Kompozy is a content engine you log into and use. This is not a feature fight; it is a "which layer of the stack are you actually shopping for" question, and the answer decides everything.
I run Kompozy, and I am not going to pretend to beat Soofi S at what it does. Soofi S is a genuinely impressive release: an open, German-built ~30B mixture-of-experts model that activates only about 3.2B parameters per token, tops fully open models on both German and English benchmarks, and was trained in Germany on sovereign infrastructure. If your reason for looking is "I want an efficient open model I can run and audit myself, with especially strong German," Soofi S is one of the best answers available right now — and Kompozy is not a substitute for it.
The split is structural. Soofi S gives you a text model and nothing else — no image or video generation, no captioning, no design layer, no scheduler, no publishing — and its first release is a base model aimed at research and further post-training, not a polished assistant. To turn it into published content you would build the whole stack around it: inference hosting, instruction tuning, media generation, brand styling, a scheduler, and integrations for every platform. Kompozy is that stack, already built and managed on Claude and OpenAI. If your bottleneck is shipping content rather than operating a model, that is the real comparison.
Everything below reconciles Soofi S against its public model card and the consortium's July 13, 2026 release materials, and Kompozy pricing against ours, both checked on 2026-07-15.
Soofi S is an open large language model released on July 13, 2026 by the Soofi project, a German research consortium coordinated by the KI Bundesverband (German AI Association) with members including Fraunhofer IAIS and IIS, the DFKI, TU Darmstadt, the University of Würzburg, the L3S Research Center, Berlin University of Applied Sciences, and the companies Ellamind and Merantix Momentum, funded by the German Federal Ministry for Economic Affairs and Energy under the European IPCEI-CIS program. Its full name is Soofi-S-30B-A3B: about 31.6 billion total parameters, but a mixture-of-experts design activating only roughly 3.2 billion per token, built on a hybrid Mamba-2/attention architecture that keeps throughput high on long inputs. It was trained on about 27 trillion tokens across three phases on Deutsche Telekom's Industrial AI Cloud in Munich, deliberately weighted toward German (from 7.2% of an early phase to 15.3% later), and reports the top aggregate score among fully open models on German and English — ahead of prior leaders like OLMo 3 32B and Apertus 70B. What it does, concretely, is generate and reason over text — drafts, translations, summaries, code — with unusually strong German. What it does not do is anything downstream of text. It is text-only: no image, video, or audio generation, no captioning, no design templates, no scheduler, no platform publishing. Those live in the application layer you build on top. You reach it by downloading the weights from Hugging Face, and because the initial release is a base model, most production use will involve fine-tuning it or waiting for an instruction-tuned variant. Note the license had not been finalized at release (a small share of training data comes from commercially licensed newspaper archives), so confirm current terms before commercial use.
The reason to look past "just use Soofi S" for a content workflow is that a raw base model is a long way from a published post. To go from Soofi S to a TikTok or a LinkedIn carousel you would host inference, fine-tune the base model into something that follows instructions cleanly, wire up image and video generation (Soofi S does neither), build brand-styling and caption rendering, write a scheduler, and integrate every platform API — a real engineering project before a single post ships, plus the GPU bill. That is the right investment for a German enterprise, a public body, or a research team whose core requirement is data sovereignty and an auditable, self-hosted model. It is the wrong investment for a creator or agency whose job is to publish this week. None of this is a knock on Soofi S. It is doing exactly what it set out to do — be an efficient, open, German-strong base layer for sovereign AI — and its MoE efficiency makes it cheaper to run than its ~30B size suggests. It just sits one or two layers below the problem most content creators have. If you want the openness, the German strength, and self-hosting, Soofi S is excellent and you should use it. If you want finished, on-brand, scheduled content across platforms, you want the layer on top — and you almost certainly do not want to build that layer yourself.
| Feature | Soofi S | Kompozy | Note |
|---|---|---|---|
| Open weights, self-hostable | Yes | No | Soofi S weights are on Hugging Face to run on your own infrastructure. Kompozy is managed SaaS, not an open model. |
| Efficient MoE (~3.2B active of ~31.6B) | Yes | N/A | Soofi S runs closer to a 3B compute cost. Kompozy is a product, not a model you host. |
| Strong German + English text | Yes | Partial | Soofi S is German-first and tops open models on both. Kompozy generates on-brand copy via Claude/OpenAI in major languages. |
| Data-sovereignty / self-audit posture | Yes | Partial | Soofi S is built in Germany for sovereign use. Kompozy is a publishing layer, not a sovereign model. |
| Ready to use without fine-tuning | Partial | Yes | Soofi S ships as a base model aimed at post-training/research. Kompozy is log-in-and-use. |
| AI text generation (captions, scripts, blogs) | Partial | Yes | Soofi S generates raw text. Kompozy writes on-brand copy governed by a Persona Brief. |
| AI image generation | No | Yes | Soofi S is text-only. Kompozy renders photo posts, carousels, quote cards, infographics. |
| AI / avatar video generation | No | Yes | Soofi S produces no media. Kompozy ships persona/avatar video, clips, marketing shorts. |
| Branded captions + design templates (HyperFrames) | No | Yes | No design layer in a raw model. Kompozy renders pixel-exact brand styling. |
| Scheduling + autopilot | No | Yes | Soofi S has no scheduler. Kompozy ships a calendar, autopilot, and review pipeline. |
| Multi-platform publishing (9 platforms + email + blog) | No | Yes | Soofi S publishes nothing. Kompozy fans output to all destinations from one queue. |
| Works without ML engineering / GPUs | No | Yes | Running and tuning Soofi S needs infra and ops. Kompozy is fully managed. |
| Tier | Soofi S plan | Soofi S price | Kompozy plan | Kompozy price |
|---|---|---|---|---|
| Entry | Soofi S (self-hosted base model) | Free weights + your own GPU/inference cost + fine-tuning | Kompozy Creator | $49/mo (2,500 credits) |
| Mid | Soofi S via a hosted inference provider | Provider inference pricing (if/when offered) | Kompozy Pro | $299/mo (18,000 credits) |
| Top | Soofi S fine-tuned / on-prem | Engineering + infra (custom) | Kompozy Enterprise | Custom (sales-led) |
Here is the honest pitch, because Soofi S and Kompozy live on different floors of the same building. Soofi S is a foundation model — and an impressive one, because it is open, efficient, German-strong, and self-hostable, with the top aggregate open-model scores on German and English. If your problem is "I need an open, sovereign model I can run and tune myself," Soofi S is a genuinely great answer and you should not be reading a Kompozy page for it.
But a model is not a content operation — and a base model even less so. To get from Soofi S to a published TikTok, Reel, carousel, or newsletter you would build everything above the model: inference hosting, instruction tuning, image and video generation (Soofi S does neither), brand styling and captions, a scheduler, and integrations for nine platforms. That is a serious engineering project plus a GPU bill. Kompozy is that entire layer, already built and managed — it generates 18 content formats across video, image, text, blog, and newsletter, holds one brand voice through a Persona Brief, and publishes to nine platforms plus email and blog on a schedule, on autopilot.
The cleanest way to think about it: if you care most about owning, auditing, and tuning the model, choose Soofi S. If you care most about producing and shipping content, choose Kompozy — and if you happen to want both, a German-sovereignty-minded team can draft in its own Soofi S deployment and let Kompozy turn those drafts into finished, scheduled posts. Start on Kompozy Creator at $49/mo (2,500 credits) to test the production half.
Not really — they sit at different layers. Soofi S is an open foundation model you download, host, and often fine-tune; Kompozy is a content generation and publishing engine you log into. People compare them because both involve AI, but Soofi S produces raw text while Kompozy produces finished, scheduled posts across platforms. For most content workflows they are complementary, not competing.
Soofi S can draft the text — with unusually strong German — but it cannot create images or video, design posts, or publish anything, and its first release is a base model rather than a ready assistant. To turn a Soofi S draft into published content you either build that pipeline yourself or use a content engine like Kompozy that generates the media and publishes to nine platforms.
When your hard requirement is data sovereignty, self-hosting, an auditable open model, or especially strong German — for example a German enterprise, public body, or research team that needs to run and tune the model on its own infrastructure. In those cases an open model like Soofi S is exactly right and a hosted SaaS is not.
Soofi S itself is free to download, but your real cost is the GPU/inference hosting, any fine-tuning, and the entire pipeline you build around it (and you should confirm the release license before commercial use). Kompozy is a managed subscription starting at $49/mo (2,500 credits) for Creator and $299/mo (18,000 credits) for Pro, with no infrastructure to run.
Yes, and for German sovereignty-minded teams that is the natural setup: draft and translate copy in your own Soofi S deployment so the text stays on your infrastructure, then bring those drafts into Kompozy to generate the video, images, and carousels and publish across platforms. Soofi S owns the open, German-first text; Kompozy owns the media and the publish.