European AI: who makes it, who runs it, and how to use it
European AI covers three things: AI models made by European companies such as Mistral AI, AI run by European operators in data centres in Europe, and the EU's AI Act, the law that sets duties for model makers, AI services and their users. Evanesca is a private AI agent and chat that makes the second one true for every model it offers: it runs 16 models in Europe — 5 of them made by European makers — and keeps no copy of your messages.
Facts checked on 6 Nov 2026
What “European AI” covers
The words are used for three different things, and a service can be European on one count and not on the others.
- Made in Europe: a European company or institute trained the model. Where it runs is a separate question — a European model can be served from the United States, and an American or Chinese open-weight model can be served from Paris.
- Run in Europe: a European company runs the model on GPUs in Europe, so your words are processed there. A US cloud's EU region passes the place test but not the company test: the CLOUD Act lets US authorities require a provider under US jurisdiction to disclose data in its “possession, custody, or control”, wherever it is stored source.
- Governed in Europe: the AI Act, Regulation (EU) 2024/1689, sets duties for the makers of general-purpose AI models, for AI services and for some uses source. With the GDPR, it is the legal frame for AI in the EU; on its own, it does not make a service European.
On this page, “Europe” means the European Union and Switzerland. Switzerland is outside the EU, but the European Commission recognises its data protection as adequate source. Open weights are what let a European operator run a model from anywhere: what open-weight AI is explains how.
European AI models: who makes them
Of the 16 models Evanesca runs, 5 come from makers based in Europe: Mistral AI (France) source, the Swiss AI Initiative (Switzerland: EPFL, ETH Zurich and CSCS) and Aleph Alpha (Heidelberg, Germany) source. Aleph Alpha says its Kolibri was “developed and trained in Europe”; it has agreed to combine with Cohere, a deal that “remains subject to regulatory approval”, and until it closes Aleph Alpha “continues to operate independently” source.
| Model | Maker | Weights | Run by, where |
|---|---|---|---|
| Mistral Large 4 | Mistral AI | Open, Mistral Ecosystem Open-Weight License | Mistral AI, in its EU region |
| Kolibri 1 | Aleph Alpha | Open, Apache-2.0 | AKI.IO, in Germany |
| Apertus v1.5 70B | the Swiss AI Initiative | Open, Apache-2.0 | Infomaniak, in Switzerland |
| Mistral Small 4 | Mistral AI | Open, Apache-2.0 | Mistral AI, in its EU region |
| Mistral Small 3.2 | Mistral AI | Open, Apache-2.0 | Scaleway, in Paris, France, with OVHcloud as a fallback |
For images, FLUX.1 [schnell] comes from Black Forest Labs (Germany) and is run by Infomaniak in Switzerland image generation. French open-source AI tells the story of Mistral AI and of the French companies around it, from model makers to the data centres that run them.
Who runs AI in Europe: operators and data centres
Every model Evanesca serves runs with one of these European operators. The company line comes from each operator's own pages; the last column is computed from Evanesca's catalogue.
| Operator | Company | Where | Runs for Evanesca |
|---|---|---|---|
| OVHcloud | OVH SAS, France — no US parent company | Gravelines, France | fallback for GPT-OSS 120B, Mistral Small 3.2 and Llama 3.3 70B |
| Scaleway | Scaleway SAS, Paris — iliad group, no US parent | Paris, France | DeepSeek V4 Flash, Qwen3.6 35B, Qwen3.5 397B, GPT-OSS 120B, Qwen3 235B, Mistral Small 3.2 and Llama 3.3 70B; fallback for DeepSeek V4.1 Flash and GLM-5.3 |
| Mistral AI | Mistral AI SAS, Paris — serves its own EU region, no US parent | Mistral AI's EU region | Mistral Large 4 and Mistral Small 4 |
| AKI.IO | AKI.IO GmbH, Berlin — a German company; its shareholders are not published | Germany | Kolibri 1, GLM-5.3, Qwen3.8 27B and Gemma 4 26B; fallback for DeepSeek V4.1 Flash |
| Verda | Verda Cloud Oy, Helsinki — a Finnish company (formerly DataCrunch), backed mainly by Nordic investors | Germany | Kimi K3.1 Pro and DeepSeek V4.1 Flash |
| Infomaniak | Infomaniak, Geneva — an independent Swiss company, no US parent | Switzerland | Apertus v1.5 70B; image generation |
OVHcloud, Scaleway, Mistral AI and Infomaniak are European companies with no US parent, out of the CLOUD Act's direct reach. AKI.IO and Verda are European companies whose shareholders are not all published, so Evanesca claims no more than that. Evanesca keeps no copy of your messages; on the operator side, Scaleway may keep and read requests that cause errors or look abusive, for up to 14 days source, while Mistral AI keeps no input or output beyond the time needed to answer: Evanesca's account has been on zero data retention since 17 November 2026 source. The sovereignty page gives each operator's statement.
Other companies run AI models in Europe too — in October 2026, IONOS, STACKIT and T-Systems in Germany, Berget AI in Sweden, Regolo in Italy and TensorX in Ireland, among others. European inference providers checks each one's company, data centres and prompt policy.
Models made elsewhere, run in Europe
The other 11 come from China (Moonshot AI, DeepSeek, Z.ai and Qwen) and the United States (Google, OpenAI and Meta). Their weights are published, so a European operator can run them without the maker: your messages go to the operator, in Europe, not to the maker. OpenAI's own GPT-OSS 120B, for instance, is run by Scaleway, in Paris, France, with OVHcloud as a fallback.
Where a model comes from still matters — for its training data, its licence and how it behaves. That is why each model's page shows six checks before you use it — licence, provenance, security, GDPR, benchmarks and a live test — and 11 of the 16 pass all six. The comparison tool and the benchmarks put them side by side.
The AI Act: Europe's law on AI
The AI Act entered into force on 1 August 2024 and applies in stages source: banned practices and AI literacy from 2 February 2025, rules for general-purpose AI models from 2 August 2025, and general application — including the transparency rules of article 50 — from 2 August 2026. Since the “Digital Omnibus on AI”, Regulation (EU) 2026/1744, the high-risk rules for uses such as employment, credit and education apply from 2 December 2027.
It shares the work out. Makers of general-purpose models document them; a maker that releases a model under a free and open-source licence, weights included, is spared two documentation duties but still owes a copyright policy and a summary of its training content, and a model with systemic risk gets no exemption (art. 53(2)). Services like Evanesca must tell you that you are talking to an AI and mark generated content in a machine-readable way (art. 50): Evanesca marks the images it generates with IPTC metadata, and marking text is in progress. Personal, non-professional use carries no duty (art. 3(4)).
The AI Act page sets out each duty, the dates and where Evanesca stands.
How to use European AI today
- In a chat. Evanesca, Mistral Vibe or Lumo by Proton: the European ChatGPT alternatives compares them. On Evanesca, the passport under each answer says which model wrote it, which operator ran it and in which country.
- For tasks. Evanesca's Agent mode writes documents and creates images, searches the web if you switch it on, works within a spending cap and never runs code; what makes an AI agent private explains what to look for.
- In your own software. The same models through Evanesca's OpenAI-compatible API, paid per token — the docs list which models run where — or straight from a European inference provider.
- On your own hardware. Download open weights and run them yourself; self-hosting vs paying per token and your own cluster or an API help you decide.
On Evanesca you start with €5 of credit, VAT included: no subscription required, and each model's price is public pricing.
What European AI does not give you yet
- The models behind ChatGPT are not published, so no European operator can run them; the same goes for other closed models. If you need one of those, you use its maker's service.
- European-made models are still a minority: 5 of the 16 models Evanesca runs.
- Evanesca's website and chat server are hosted by Vercel, a US company, in its Paris region; the models run with the European operators above privacy.
- Dictation goes to Mistral AI, a French company, but that call is not pinned to its EU region.
Frequently asked questions
What is European AI?
It means three things: AI models made by European companies, such as Mistral AI's; AI run by European companies in data centres in Europe, so your data is processed there; and the EU's AI Act, which sets rules for model makers, AI services and some uses. A service can be European on one count and not on the others.
Which AI models are made in Europe?
Among the models Evanesca runs: Mistral Large 4 (Mistral AI), Kolibri 1 (Aleph Alpha), Apertus v1.5 70B (the Swiss AI Initiative), Mistral Small 4 (Mistral AI) and Mistral Small 3.2 (Mistral AI). For images, FLUX.1 [schnell] comes from Black Forest Labs (Germany).
Is Mistral AI a European company?
Yes: Mistral AI is a French company registered in Paris. Evanesca runs Mistral Large 4, Mistral Small 4 and Mistral Small 3.2, either with Mistral AI in its EU region or with a French operator.
Is an AI hosted in an EU data centre European?
Not necessarily. If the provider is a US company, the CLOUD Act lets US authorities require it to disclose data it controls, wherever it is stored. A European AI service needs a European company as well as a European place.
Does the AI Act make an AI model European?
No. The AI Act sets duties — documentation for model makers, transparency for services, stricter rules for high-risk uses — Who made a model and where it runs are separate questions, and they are what make AI European.
Read next
- The best ChatGPT alternatives, by what you needThe best ChatGPT alternative depends on your need: privacy, European law, free use, research, code or offline. Claude, Lumo, Duck.ai, Vibe and more.
- Private AI agent: what makes an agent private, and how Evanesca's worksA private AI agent carries out tasks without keeping your data. What to check — where it runs, what it keeps, what it can touch — and Evanesca's Agent mode.
- Private AI: what it means, and how to judge onePrivate AI keeps your words from being stored, read or used for training. The three kinds, a checklist to judge any of them, and the main options compared.
- Is there a European ChatGPT? What it means, and the optionsIs there a European ChatGPT? What to check — EU company, servers in Europe, the CLOUD Act — and how Mistral Vibe, Lumo and Evanesca compare.
- Sovereign AI: what the word covers, and how to check a claimSovereign AI means control of the model, the servers, the operator and the law. A checklist to judge any claim, and where Evanesca stands on each layer.
- AI for lawyers and law firms: what it is good for, and how to keep client secretsAI for lawyers: drafting, summaries, translation, contract review — and client confidentiality. What the French bar says, a checklist for law firms.