Apertus 70B
Apertus 70B is an open-weights language model from ETH Zurich, released September 2, 2025. It has 71B parameters (dense). The context window is 64K tokens.
Researchers at EPFL, ETH Zurich and CSCS have developed Apertus, a fully open Large Language Model (LLM) – one of the largest of its kind. As a foundational technology, Apertus enables innovation and strengthens AI expertise across research, society and industry by allowing others to build upon it.
Check it against your own PC →
Can you run Apertus 70B locally?
The smallest common GPU it runs on is 24 GB of VRAM (with 32 GB of RAM): IQ2_XXS, split GPU / CPU, roughly 15 tokens per second. Estimates assume 32 GB of DDR5 RAM, a 16K context and llama.cpp-style GGUF quantization. Enter your exact hardware in the live app.
| GPU memory | Verdict | Best quant | VRAM | RAM | tok/s |
|---|---|---|---|---|---|
| 6 GB VRAM | Too large | – | – | – | – |
| 8 GB VRAM | Very slow | IQ2_XXS (heavy loss) | 7.2 GB | 16.9 GB | ~2 |
| 12 GB VRAM | Very slow | IQ2_XXS (heavy loss) | 11.2 GB | 12.9 GB | ~2 |
| 16 GB VRAM | Very slow | IQ2_XXS (heavy loss) | 15.2 GB | 8.9 GB | ~3 |
| 24 GB VRAM | Runs, slowly | IQ2_XXS (heavy loss) | 23.2 GB | 0.9 GB | ~15 |
| 32 GB VRAM | Runs on GPU | IQ2_M (heavy loss) | 29.8 GB | – | ~43 |
Ready-made quantizations on Hugging Face: from-our-page/Apertus-70B-2509-GGUF · bg-digitalservices/Apertus-70B-2509-NVFP4 · bg-digitalservices/Apertus-70B-2509-NVFP4A16 · unsloth/Apertus-70B-Instruct-2509-unsloth-bnb-4bit
Benchmark results
Epoch AI has not published results for Apertus 70B yet. New models are usually evaluated within days to weeks.
Links
Announcement · Hugging Face: swiss-ai/Apertus-70B-2509