Apertus 70B
Apertus 70B comes from Switzerland, developed by EPFL, ETH Zurich and the Swiss National Supercomputing Centre CSCS. Unlike most models, the training data and training process are published too, and the model was trained on text in more than 1,000 languages. There are no independent ratings from LMArena or Epoch yet, so test it with your own tasks before use.
Profile
- Maker
- Swiss AI (EPFL, ETH Zürich, CSCS) · Switzerland
- Licence
- Free for commercial use
- Runs on
- Mac Studio with 128 GB
- Memory at 4 bits
- 50 GB (estimated)
- Parameters
- 72bn
- Context
- 100,000 tokens
- Inputs
- text, image, audio
- Released
- July 2026
Scores
No independent scores from LMArena or Epoch AI yet.
Scores: LMArena, Epoch AI, EuroEval, Hugging Face, models.dev; as of 19 October 2026. Sources and method →
Use cases
Where the model makes sense.
Public administration and education
- Situation
- An authority, university or school wants to use AI and must be able to show what the model was trained on.
- What it looks like
- Apertus runs in your own data centre; training data and licence are publicly documented and can be carried over into data protection and AI documentation.
- What to watch
- Open training data makes documentation easier but does not replace assessing the specific use.
Multilingual content, including rare languages
- Situation
- A company or organisation communicates with people in many languages, including smaller ones.
- What it looks like
- The model drafts texts and translations in languages many other models barely cover; native speakers review them.
- What to watch
- Quality varies a lot by language. Test each language separately.
Data sovereignty in Switzerland
- Situation
- A Swiss company wants a model from Switzerland running on its own infrastructure.
- What it looks like
- Apertus runs on your own hardware; if you do not want to run a server, you can use it via Lumo from Geneva-based Proton.
- What to watch
- With Lumo, Proton’s terms apply; with your own operation, your own.
AI research and teaching
- Situation
- Universities want to research and teach how language models are built and behave.
- What it looks like
- Because data and training are open, experiments can be reproduced and results traced back to the training data.
- What to watch
- For comparisons, also use a fully open model such as Olmo.
The use cases were written with AI assistance (Claude by Anthropic) and reviewed by vona. Scores come unchanged from the sources listed.
When another model fits better
Where performance in coding, maths or agents matters most, Qwen or GLM models in the same hardware class score higher on the available measurements.
How to start
The version 1.5 weights are on Hugging Face and run with programs such as vLLM; a quantised GGUF version for LM Studio or Ollama so far exists only for the older version. Without your own hardware you can try Apertus via Proton’s Lumo.
Suitable interfaces and programs from our directory: LibreChat, Lumo
Licence: Free for commercial use
The licence allows commercial use, modification and redistribution; the main obligation is to keep the licence notice.
Hugging Face ↗Apertus as a cloud service in the AI tools directory →Similar models
Comparable models.
Swiss AI (EPFL, ETH Zürich, CSCS) · Switzerland
Apertus 8B
Small version of Apertus for laptops, also with open training data.
Free for commercial use · Runs on: Laptop with 16 GB
In the overview →Allen Institute for AI · United States
Olmo 3.1 32B
Fully open model from the Allen Institute for AI: training data and training code are published alongside the weights. Useful when you need to trace what a model was trained on.
Free for commercial use · Runs on: Computer with 32 GB
In the overview →UTTER-Konsortium (EU) · European Union
EuroLLM 22B
Model developed by a European research consortium that covers all 24 official EU languages.
Free for commercial use · Runs on: Computer with 32 GB
In the overview →
FAQ
What hardware do I need for Apertus 70B?
In the common 4-bit version, Apertus 70B needs about 50 GB of memory including headroom for context and runtime. That fits: Mac Studio with 128 GB. Long documents increase the requirement.
May I use Apertus 70B commercially?
The licence allows commercial use, modification and redistribution; the main obligation is to keep the licence notice.
How good is Apertus 70B in German?
There are no independent German scores for Apertus 70B yet. Test it with your own examples before using it.
AI consulting
Your own language model for your company?
Together we find the model that fits your tasks, your hardware and your data protection requirements, test it with real documents and set up operation.