// OPEN FOUNDATION MODEL / LLM REVIEW

Apertus Review (2026): Honest Verdict on Switzerland's Fully Open Sovereign AI Model

A working review of Apertus, the fully open Swiss LLM from EPFL, ETH Zurich, and CSCS. What it nails on openness and languages, where it stops, and who it fits.

Last verified · 2026-06-22 · by Moe Ameen
The verdict
4.3 / 5

Apertus is one of the most genuinely open large language models ever shipped — weights, training data, code, and alignment recipe all public, under Apache 2.0, built by Swiss public institutions for sovereign and EU-compliant AI. Judged as what it is, an open multilingual foundation model, it is excellent and important. It is not a content tool: there is no image, video, captioning, or publishing layer, and running it well takes real infrastructure. Score it high for openness and multilingual reach, and look elsewhere if you came to produce and ship content.

Most coverage of Apertus is either a flag-waving "Europe finally has its own model" piece or a benchmark table that misses the point. This review is neither. We build a content engine and we read model cards for a living, so the goal here is to tell you what Apertus is genuinely good at, where it stops, and — because people arrive at this question sideways — whether it can power a content operation.

Short version up top: Apertus is a landmark open model. Released on September 2, 2025 by the Swiss AI Initiative (EPFL, ETH Zurich, and the Swiss National Supercomputing Centre), it is "fully open" in the strict sense — not just downloadable weights, but the training data, source code, training recipe, and alignment principles, all published and reproducible under Apache 2.0. For sovereignty, auditability, and broad multilingual coverage, that combination is rare and valuable.

The honest catch is scope and operability. Apertus is a text-and-reasoning model. It generates and translates copy across more than 1,000 languages, but it renders no images, no video, no audio, and it publishes nothing. And running the 70B variant well is an infrastructure project, not a download-and-go. None of that is a flaw — it set out to be an open base layer, not a finished application — but it is the thing to understand before you decide it fits your workflow.

This review covers what Apertus actually is in 2026, how its openness and licensing stack up, where it is strong, where it is honestly the wrong tool, and who should use it versus who should keep looking.

What Apertus is

Apertus is a fully open large language model from the Swiss AI Initiative. It ships in two sizes — 8 billion and 70 billion parameters — each with a pretrained base and an instruction-tuned variant, was trained on roughly 15 trillion tokens across more than 1,000 languages (about 40% non-English), and supports a long 65,536-token context. Architecturally it is a decoder-only transformer using a novel xIELU activation and the AdEMAMix optimizer, trained on thousands of GH200 GPUs at CSCS. It is released under the permissive Apache 2.0 license, which allows research, education, and commercial use. What sets it apart is transparency. Where most "open" releases ship a checkpoint and little else, Apertus publishes the training data, code, methods, and alignment recipe, and even intermediate checkpoints — the whole pipeline is documented and reproducible. The corpus was built only from publicly available data, filtered to respect machine-readable opt-outs (even retroactively) and to strip personal data, with EU AI Act transparency and Swiss copyright and data-protection law in mind. You can download it from Hugging Face (the swiss-ai org), run it locally via tools like LM Studio, or reach it through partners such as Swisscom and the Public AI Inference Utility.

Who Apertus is for

The clearest fit is anyone whose requirement is sovereignty or transparency: public institutions, research labs, and regulated enterprises that need to run and audit a model on their own infrastructure, plus builders who want a permissively licensed base model to fine-tune without vendor lock-in. It is also a strong choice for teams serving multilingual audiences, including under-served languages like Swiss German and Romansh. It is the wrong tool for someone whose actual output is published content — video, images, carousels, social posts — because producing and distributing that content is entirely outside what a text model does, and for non-technical users who want a hosted, log-in-and-go experience rather than a model to operate.

Scoring breakdown

DimensionScoreWhy
Openness & transparency5.0 / 5Weights, training data, code, methods, alignment, and intermediate checkpoints all published. About as open as a model at this scale gets.
License & commercial freedom4.8 / 5Apache 2.0 — commercial use allowed with no per-seat or per-token fee to the model itself. Very few peers at this scale match it.
Multilingual coverage4.7 / 5Trained across 1,000+ languages with heavy non-English weighting, including languages most models ignore. A genuine differentiator.
Data compliance & sovereignty posture4.7 / 5Built around EU AI Act transparency and Swiss data law; corpus respects opt-outs and strips personal data. Strong for regulated and public-sector use.
Model quality & reasoning4.0 / 5Competitive with strong open models and credible against closed ones, but not positioned as a frontier-leading reasoner. Solid, not record-setting.
Ease of use / accessibility3.3 / 5Downloadable and partner-hosted, but running the 70B well needs real infra and ML ops. Not a non-technical, log-in-and-go experience.
Ecosystem & tooling3.6 / 5Available on Hugging Face, LM Studio, Swisscom, and Public AI, with active docs — a growing but still young ecosystem versus the big closed providers.
Content / social media production1.0 / 5Not the product. No image, video, audio, captions, or design output. Out of scope by design.
Multi-platform publishing1.0 / 5Apertus produces text; it does not post. There is no scheduler and no platform integration.

Pros and cons

Pros

  • Genuinely fully open — weights, training data, code, methods, and alignment recipe are all published and reproducible.
  • Apache 2.0 license permits commercial use with no fee to the model itself.
  • Self-hostable, so sensitive data can stay entirely on your own infrastructure.
  • Exceptional multilingual breadth, including under-served languages like Swiss German and Romansh.
  • Built around EU AI Act transparency and Swiss data-protection law — a strong story for regulated and public-sector use.
  • Backed by credible public institutions (EPFL, ETH Zurich, CSCS) rather than a single vendor.
  • Two sizes (8B and 70B) and a long 65,536-token context cover different hardware budgets.

Cons

  • Text-only — no image, video, audio, captioning, or design output of any kind.
  • No publishing, scheduling, or platform integration; it is a model, not a content tool.
  • Running the 70B variant well requires ML skills and meaningful GPU budget.
  • Like any LLM, it can produce inaccurate or biased text that needs human review before shipping.
  • No brand-voice governance or per-post review workflow — that is all on you to build.
  • Ecosystem and hosted-tooling maturity still trails the big closed providers.

Pricing analysis

Apertus has no license price. The weights are free under Apache 2.0, so the cost question is really "what does it cost to run." Self-hosting means GPU and inference infrastructure — modest for the 8B model, substantial for the 70B at any real throughput — plus the engineering to operate it. If you would rather not run hardware, partners such as Swisscom and the Public AI Inference Utility offer hosted access at their own inference pricing, which trades the GPU bill for a per-use fee.

For the sovereignty and research use cases Apertus targets, that economic model is exactly right: a free, auditable, self-hostable model is enormous value when control and transparency are the requirement, and Apache 2.0 removes the per-seat and per-token drag of closed APIs. The catch is that "free model" is not "free outcome" — the total cost is the infrastructure and the application you build around it.

The honest framing on value is that Apertus is priced like what it is: open infrastructure. It is not priced or built as a content-marketing tool, and no amount of inference budget adds rendering or publishing. If your spend is meant to produce and distribute content, you are comparing the wrong line item.

Use-case fit

Use caseFitWhy
Sovereign / on-prem deployment for regulated or public-sector useStrongSelf-hostable and fully auditable, built around EU AI Act and Swiss-law compliance — squarely its purpose.
Multilingual text generation, including under-served languagesStrongTrained on 1,000+ languages with heavy non-English weighting; covers languages most models handle poorly.
Fine-tuning a permissively licensed base modelStrongApache 2.0 weights plus published recipe make it an ideal foundation to build on without lock-in.
Drafting copy, scripts, and summariesOKCapable text generation, but output needs a human pass and it is one input to a content workflow, not the whole thing.
Producing short-form or avatar video for socialWeakNo video generation of any kind. Entirely outside Apertus's scope.
Brand-consistent content across formatsWeakNo persona or brand-voice system, no design layer. It generates text but does not govern a content voice or render media.
Scheduling and publishing across platformsWeakNo publishing layer and no scheduler. Apertus produces text, not posts.
Non-technical, hosted, log-in-and-go useOKPartner-hosted access exists, but the experience is a model endpoint, not a finished application.

Alternatives worth considering

  • Llama and other open-weight models — comparable open options if full data transparency and sovereignty are not hard requirements.
  • Mistral models — strong open European models for those who want capable open weights with a lighter compliance story.
  • Closed APIs (Claude, GPT) — higher convenience and frontier reasoning, at the cost of openness and self-hosting.
  • Kompozy — different category entirely: a content generation and publishing engine for video, images, text, blogs, and newsletters across nine platforms.

How Kompozy compares

If you arrived at this review wondering whether Apertus can run your content operation, the honest answer is no — and that is a category question, not a criticism. Apertus is a foundation model: open, auditable, multilingual infrastructure. It has no renderer, no caption engine, no design layer, and no scheduler, because it was never meant to be a content tool. Scoring it as one would be unfair to a model that is genuinely excellent at its actual job.

Kompozy sits at the layer above. Where Apertus stops at text, Kompozy turns an idea or a draft into 18 content formats — persona and avatar video, carousels, quote cards, infographics, blogs, newsletters, and platform-native posts — holds one brand voice through a Persona Brief, and schedules and publishes across nine platforms plus email and blog. It runs that generation on managed Claude and OpenAI models, so there are no GPUs to operate. The two are not rivals: a sovereignty-minded team could draft in its own Apertus deployment and let Kompozy produce and ship the finished content. Use Apertus for the open model layer it is built for, and a content engine for the content.

Frequently asked questions

What is Apertus?

Apertus is a fully open, multilingual large language model built by the Swiss AI Initiative (EPFL, ETH Zurich, and CSCS) and released on September 2, 2025. It comes in 8B and 70B sizes, was trained on about 15 trillion tokens across 1,000+ languages, and publishes its weights, training data, code, and methods openly under the Apache 2.0 license.

Is Apertus worth it in 2026?

For sovereignty, transparency, and multilingual reach — yes, it is one of the strongest fully open models available, and free under Apache 2.0. It is not worth adopting for content production, because it generates no media and publishes nothing; for that you need a content engine on top.

How good is Apertus compared to closed models like GPT or Claude?

It is competitive with strong open models and credible against closed ones for many tasks, with standout multilingual coverage. It is not positioned as a frontier-leading reasoner, so for the hardest reasoning or coding tasks the top closed models still lead — but you trade away openness and self-hosting to use them.

How much does Apertus cost?

The model is free under Apache 2.0. Your real cost is the GPU/inference infrastructure to self-host it (modest for 8B, substantial for 70B) or a partner provider's inference pricing through services like Swisscom or the Public AI Inference Utility.

Can Apertus create social media videos or images?

No. Apertus is a text-and-reasoning model. It does not render video, generate images, write captions, or post to any platform. To turn its text into published content you need a content engine like Kompozy.

Why is Apertus called a "sovereign AI" model?

Because its entire pipeline is open and reproducible and it was built around EU AI Act transparency and Swiss data-protection law, governments, public institutions, and companies can run and audit it on their own infrastructure rather than depending on a closed foreign provider — that independence is the sovereignty argument.

Where can I download or use Apertus?

The weights are on Hugging Face under the swiss-ai organization, and you can run it locally with tools like LM Studio or access it through partners such as Swisscom and the Public AI Inference Utility.

Apertus or Kompozy for content?

Kompozy, without question. Apertus produces text and nothing else; Kompozy generates video, images, carousels, blogs, and newsletters and publishes them across platforms. Use Apertus as an open model layer — for sovereign teams, even as a drafting source — and Kompozy to produce and ship the content.

Related deep guides

See Apertus vs Kompozy comparison → · Get Started →