Apertus-70B-Instruct-2509 needs about 46.9 GB of memory at Q4_K_M (a 43.7 GB file) with an 8K context window, so it fits on a 48 GB GPU. This repository has 26 quants, from 16.6 GB (IQ1_S) to 141.2 GB (BF16). Longer context needs more: at 32K tokens the KV cache alone adds about 10.7 GB.
Apertus‑70B‑Instruct‑2509 – a 70B open‑source instruction model
Apertus‑70B‑Instruct‑2509 is a 70.6B parameter decoder‑only transformer. It supports up to 65,536 context tokens and is fully open with Apache‑2.0 license. The model is trained on 15T tokens with a staged curriculum and includes alignment via QRPO.
Good at
- Massive multilingual support – 1811 languages
- Long‑context handling – 65,536 tokens
- Instruction‑following with alignment
Limits
- Requires large VRAM – 43–86 GB depending on quant
- No vision or multimodal capabilities
- No official benchmarks listed
Best for
- Running on high‑end GPUs with 48 GB+ VRAM
- Fine‑tuning for multilingual instruction tasks
- Deploying on‑device with MLX for large‑context workloads
Who it’s for. Developers and researchers needing a fully open, multilingual instruction model with long‑context support. The model is suitable for those who can provide the necessary GPU memory and want to experiment with different quantizations.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| IQ1_S | 16.6 GB | 19.8 GB | Too big | Get file → | |
| IQ1_M | 17.8 GB | 21.0 GB | Too big | Get file → | |
| IQ2_XXS | 19.9 GB | 23.1 GB | Too big | Get file → | |
| IQ2_M | 24.8 GB | 28.0 GB | Too big | Get file → | |
| Q2_K | 27.3 GB | 30.5 GB | Too big | Get file → | |
| Q2_K_L | 27.5 GB | 30.7 GB | Too big | Get file → | |
| IQ3_XXS | 27.9 GB | 31.1 GB | Too big | Get file → | |
| Q2_K_XL | 28.2 GB | 31.4 GB | Too big | Get file → | |
| Q3_K_S | 30.8 GB | 34.0 GB | Too big | Get file → | |
| Q3_K_M | 35.5 GB | 38.7 GB | Too big | Get file → | |
| Q3_K_XL | 36.6 GB | 39.8 GB | Too big | Get file → | |
| IQ4_XS | 38.0 GB | 41.2 GB | Too big | Get file → | |
| IQ4_NL | 40.1 GB | 43.3 GB | Too big | Get file → | |
| Q4_0 | 40.2 GB | 43.4 GB | Too big | Get file → | |
| Q4_K_S | 40.4 GB | 43.6 GB | Too big | Get file → | |
| Q4_K_M | 43.7 GB | 46.9 GB | Too big | Get file → | |
| Q4_K_XL | 44.2 GB | 47.4 GB | Too big | Get file → | |
| Q4_1 | 44.3 GB | 47.5 GB | Too big | Get file → | |
| Q5_K_S | 48.7 GB | 51.9 GB | Too big | Get file → | |
| Q5_K_M | 50.6 GB (2 parts) | 53.8 GB | Too big | Get file → | |
| Q5_K_XL | 51.0 GB (2 parts) | 54.2 GB | Too big | Get file → | |
| Q6_K | 57.9 GB (2 parts) | 61.1 GB | Too big | Get file → | |
| Q6_K_XL | 62.1 GB (2 parts) | 65.3 GB | Too big | Get file → | |
| Q8_0 | 75.0 GB (2 parts) | 78.2 GB | Too big | Get file → | |
| Q8_K_XL | 83.5 GB (2 parts) | 86.7 GB | Too big | Get file → | |
| BF16 | 141.2 GB (3 parts) | 144.4 GB | Too big | Get file → |
For 48 GB VRAM, use Q4_K_M (43.72 GB file, ~46.9 GB memory) or Q4_K_XL (44.21 GB file, ~47.4 GB memory). For 64 GB VRAM, Q5_K_M (50.61 GB file, ~53.8 GB memory) or Q5_K_XL (50.97 GB file, ~54.2 GB memory) are suitable. Lower‑memory options like IQ1_S (16.6 GB file, ~19.8 GB memory) can be used on 8–12 GB VRAM systems.
What GPU do you need for Apertus-70B-Instruct-2509?
These figures add the file size, an fp16 KV cache at an 8K context window (worked out from this model’s own header) and about 0.5 GB for the runtime. They are estimates, so leave a little headroom, and expect to use less context or a smaller quant if you are right at the limit.
- 4 GB VRAM: no quant fits at an 8K context.
- 6 GB VRAM: no quant fits at an 8K context.
- 8 GB VRAM: no quant fits at an 8K context.
- 12 GB VRAM: no quant fits at an 8K context.
- 16 GB VRAM: no quant fits at an 8K context.
- 24 GB VRAM: the best quant that fits is IQ2_XXS (19.9 GB file, about 23.1 GB in use).
- 32 GB VRAM: the best quant that fits is Q2_K_XL (28.2 GB file, about 31.4 GB in use).
- 48 GB VRAM: the best quant that fits is Q4_1 (44.3 GB file, about 47.5 GB in use).
Which Apertus-70B-Instruct-2509 quant should you choose?
Q4_K_M is the usual starting point for Apertus-70B-Instruct-2509: it balances file size and quality well. Here it is a 43.7 GB file.
The Q8_0 file is 75.0 GB, about 1.7 times the size of Q4_K_M. It keeps the weights closer to the original, which only matters if you have the memory to spare.
If memory is tight, the smallest file is IQ1_S at 16.6 GB, with the quality trade-offs that come with fewer bits. Use it only when memory is very tight. Expect a clear drop in quality, which is usually worse on small models and easier to live with on large ones.
New to quant names? Read GGUF quantization explained.
Specifications
Architecture from the GGUF header
Source
Frequently asked questions about Apertus-70B-Instruct-2509
How much VRAM does Apertus-70B-Instruct-2509 need?
About 46.9 GB at Q4_K_M with an 8K context window. That is the 43.7 GB file, plus the KV cache and roughly 0.5 GB of runtime overhead. Smaller quants need less and longer context needs more.
Can I run Apertus-70B-Instruct-2509 on an 8 GB GPU?
Not at an 8K context. The smallest file needs about 19.8 GB, so you would need a larger GPU or CPU offloading.
What is the file size of Apertus-70B-Instruct-2509 GGUF?
It depends on the quant: IQ1_S is 16.6 GB, IQ1_M is 17.8 GB, IQ2_XXS is 19.9 GB, IQ2_M is 24.8 GB, Q2_K is 27.3 GB, Q2_K_L is 27.5 GB and 20 more.
What context length does Apertus-70B-Instruct-2509 support?
The GGUF header lists a trained context length of 65,536 tokens (64K). The KV cache grows with the context you use: about 2.7 GB at 8K and 10.7 GB at 32K.
Where can I download Apertus-70B-Instruct-2509 GGUF?
The files are on Hugging Face in the unsloth/Apertus-70B-Instruct-2509-GGUF repository. The download buttons on this page open the file on Hugging Face. ggufmodels.com does not host any model files.
What license does Apertus-70B-Instruct-2509 use?
Hugging Face lists the license as apache-2.0. Licenses can differ between the quantized files and the original model, so check the repository and the base model before commercial use.
Can I run Apertus‑70B‑Instruct‑2509 on 8 GB VRAM?
No, the smallest quant requires 19.8 GB memory at 8k context. 8 GB VRAM is insufficient.
Does the model support vision or multimodal inputs?
No, the model is a decoder‑only language model with no vision capabilities.
What license applies to this model?
The model is released under the Apache‑2.0 license.