Apertus 70B
Swiss AI Initiative·
LLMsopen-weightWorkstation
Overview
Fully open, transparent, multilingual model family (8B and 70B) from the Swiss AI Initiative (EPFL, ETH Zurich, CSCS). Decoder-only dense transformer pretrained on ~15T tokens with a staged web/code/math curriculum, supporting 1000+ languages, long context, and using only compliant, fully open training data. Weights, data and training recipe are released. Distributed via Hugging Face, Swisscom and the Public AI network.
Capabilities and innovations
1000+ languages (sovereign Swiss/EU model)Fully open weights, data and training recipeLong-context support8B variant runs on consumer GPUsOne of the most transparent open models: compliant, fully open training dataMassively multilingual coverage (1000+ languages)Trained on Swiss national supercomputer (CSCS Alps)
Benchmarks
No benchmark values are recorded for this model.
Architecture and hardware
- Parameters
- 70B total (Dense)
- Estimated VRAM at Q4
- ~41 GB, Workstation class
- Quantization formats
- GGUF
- Recommended runtime
- Ollama
- License
- Apache 2.0
Links
Data curated by AI Model Timeline. See the methodology for admission criteria, benchmark eras and source priority.