All models › swiss-ai
Apertus-v1.5-8B
swiss-ai · license to accept license · on Arena since 24 Jul 2026
Download No Arena votes
Arena score—Not voted on Arena yet
Size8.9 Bparameters
4-bit download5.4 GBweights only, Q4_K_M
Max context—
Where to use it for free
- Download and run it at home: official weights · GGUF files for Ollama and LM Studio (you accept the license on Hugging Face first)
No Arena votes yet
Arena has not voted on this model. It is listed because it is a popular small model on its lab's official Hugging Face page (556K downloads a month).
Speed on popular hardware
| Hardware | GPU memory | Fits? | Speed |
|---|---|---|---|
| RTX 3060 12 GB | 12 GB | Fits in GPU | 38 tok/s |
| RTX 4060 Ti 16 GB | 16 GB | Fits in GPU | 31 tok/s |
| RTX 3090 | 24 GB | Fits in GPU | 100 tok/s |
| RTX 4090 | 24 GB | Fits in GPU | 108 tok/s |
| RTX 5090 | 32 GB | Fits in GPU | 191 tok/s |
| RX 7900 XTX | 24 GB | Fits in GPU | 102 tok/s |
| 2 × RTX 3090 | 48 GB | Fits in GPU | 97 tok/s |
| 4 × RTX 3090 | 96 GB | Fits in GPU | 92 tok/s |
| Mac M4 Max 128 GB | 96 GB | Fits in GPU | 58 tok/s |
| Ryzen AI Max+ 395 128 GB | 96 GB | Fits in GPU | 27 tok/s |
| 8 × H100 SXM | 640 GB | Fits in GPU | 295 tok/s |
Estimate at 4-bit with 32K context, 64 GB of DDR5 for offload. Speed ≈ memory bandwidth × 70% ÷ data read per token; real numbers vary by about ±30%.
GPU memory needed
| Level | 8K context | 32K context | 128K context | Quality vs full |
|---|---|---|---|---|
| Q3_K_M | 5.3 GB | 6.2 GB | 9.7 GB | +10.2% noticeable loss |
| Q4_K_M | 6.3 GB | 7.2 GB | 11 GB | +2.7% small loss |
| Q5_K_M | 7.2 GB | 8.1 GB | 12 GB | +0.9% very small loss |
| Q6_K | 8.2 GB | 9.1 GB | 13 GB | +0.3% near-lossless |
| Q8_0 | 10 GB | 11 GB | 15 GB | ≈ 0% same as full |
| F16 | 19 GB | 20 GB | 23 GB | ≈ 0% same as full |
Weights + context cache (from the model's architecture) + runtime. Quality: increase in perplexity against the full model, from llama.cpp tests, scaled by model size. Simulate your own setup.