Open-weight AI models made in Europe
Developers in France, Germany, Switzerland and the Netherlands release open-weight models for text, images, speech, documents and search. This list gathers the current ones in our catalog and notes where a license is less open than the label suggests.
-
Mistral Large 4
Mistral's roughly 1T-parameter multimodal mixture of experts with 52B active parameters and a 1M context. It uses a modified MIT license, and businesses that sell hosted access need a commercial license from Mistral.
-
Apertus 1.5
Switzerland's fully open 8B and 70B models: the weights, data preparation, training code and evaluations are all public, and they read images and audio as well as text. Apache 2.0.
-
Kolibri 1
Aleph Alpha's English-German reasoning model: a 78B mixture of experts with only 3.5B active parameters, a 1M context and the Apache 2.0 license. By our estimate it needs about 54 GB at 4-bit.
-
Mellum2.1
JetBrains' 12B mixture of experts with 2.5B active parameters for coding agents, under Apache 2.0. JetBrains was founded in Prague and is headquartered in Amsterdam.
-
FLUX 3 Image Klein
Black Forest Labs' 5.6B model generates and edits images, and an INT8 build runs in 8 GB of VRAM. Commercial use is free only below US$5 million of annual revenue, and the download is gated.
-
P-H3-Image
Pruna AI's 8-step image and editing model, distilled from MiniMax-H3 by a Munich-based company. Despite its origin, the MiniMax H3 license excludes the European Union, the UK, South Korea and the United States, so check where you can use it.
-
Voxtral Realtime
Mistral's Apache 2.0 4B streaming speech-to-text model: 13 languages and a delay you set between 80 ms and 2.4 s.
-
Voxtral TTS
Mistral's 4B open-weights text-to-speech model: 9 languages and voice adaptation from a short clip. The CC BY-NC license makes it non-commercial only.
-
Pocket TTS
Kyutai's 100M voice-cloning speech model from its Paris lab, which runs faster than real time on a laptop CPU. The weights are CC BY 4.0 behind a prohibited-use gate.
-
Phonon 31M
Gradium's 31M English speech model in a 64.5 MB file under Apache 2.0, from the Paris company spun out of Kyutai. Its larger multilingual models are non-commercial.
-
Phonon 40M
Gradium's 40M speech models for English, French, German, Spanish and Portuguese, including a multilingual one, in 74.6 MB files. They are CC BY-NC and gated, so not free for commercial use.
-
LightOnOCR-3
LightOn's Apache 2.0 OCR family in 4B, 1B and 0.8B sizes. It writes Markdown and can also return bounding boxes, image captions and chart data.
-
Jina Embeddings v5
Jina AI's v5 pair of 677M and 239M multilingual embedding models, distilled from Qwen3-Embedding-4B. They are free for non-commercial use only under CC BY-NC.
-
Jina Embeddings v5 Omni
Jina AI's small and nano models that put text, images, video and audio in one space, keeping text vectors identical to v5 text. They are CC BY-NC, so a commercial product needs a separate license.
-
Jina Reranker v3.5
Jina AI's 0.6B listwise reranker that scores a whole candidate list in one pass, with 131K context. It is CC BY-NC, so a commercial product needs a separate license from Jina.
How we chose
- The developer is based in Europe, which in our catalog means France, Germany, Switzerland or the Netherlands, according to the country on its developer page.
- Open weights you can download today from the developer's official repository.
- Current releases only, and the license of each is shown, because several of these models are free for non-commercial use only, only below a revenue limit, or not in every region.