Qwen3.5 35B A3B needs about 23.5 GB of memory at Q4_K_M (a 22.3 GB file) with an 8K context window, so it fits on a 24 GB GPU. This repository has 29 quants, from 0.2 GB to 71.1 GB (BF16). Longer context needs more: at 32K tokens the KV cache alone adds about 2.8 GB.
Qwen3.5 35B A3B – 35B parameter vision‑language model
Qwen3.5 35B A3B is a 35‑billion‑parameter model with 3 billion active experts. It supports vision and text, trained on 262,144‑token context. The model is released under Apache‑2.0 and is compatible with llama.cpp and other inference engines.
Good at
- Vision‑language reasoning
- Large‑context text generation
- Multilingual support
Limits
- Requires high VRAM for full precision
- No built‑in vision tools in llama.cpp
- License allows open use but no commercial restrictions
Best for
- Multimodal chatbots
- Long‑form content creation
- Research on vision‑language models
Who it’s for. Developers and researchers needing a large vision‑language foundation model. The model is suitable for those who can allocate sufficient GPU memory and want to experiment with Mixture‑of‑Experts architectures.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| — | 0.2 GB | 1.4 GB | Fits | Get file → | |
| IQ1_M | 9.4 GB | 10.6 GB | Too big | Get file → | |
| IQ2_XXS | 10.7 GB | 11.9 GB | Too big | Get file → | |
| IQ2_XS | 11.7 GB | 12.9 GB | Too big | Get file → | |
| IQ2_S | 11.9 GB | 13.1 GB | Too big | Get file → | |
| IQ2_M | 13.0 GB | 14.2 GB | Too big | Get file → | |
| Q2_K | 13.5 GB | 14.7 GB | Too big | Get file → | |
| Q2_K_L | 14.0 GB | 15.2 GB | Too big | Get file → | |
| IQ3_XXS | 15.8 GB | 16.9 GB | Too big | Get file → | |
| Q3_K_S | 16.4 GB | 17.6 GB | Too big | Get file → | |
| IQ3_XS | 17.1 GB | 18.3 GB | Too big | Get file → | |
| Q3_K_M | 17.1 GB | 18.3 GB | Too big | Get file → | |
| Q3_K_L | 17.8 GB | 19.0 GB | Too big | Get file → | |
| IQ3_M | 17.8 GB | 19.0 GB | Too big | Get file → | |
| Q3_K_XL | 18.2 GB | 19.4 GB | Too big | Get file → | |
| IQ4_XS | 19.7 GB | 20.9 GB | Too big | Get file → | |
| IQ4_NL | 20.8 GB | 21.9 GB | Too big | Get file → | |
| Q4_0 | 20.8 GB | 22.0 GB | Too big | Get file → | |
| Q4_K_S | 21.5 GB | 22.7 GB | Too big | Get file → | |
| Q4_K_M | 22.3 GB | 23.5 GB | Too big | Get file → | |
| Q4_K_L | 22.7 GB | 23.9 GB | Too big | Get file → | |
| Q4_1 | 22.9 GB | 24.1 GB | Too big | Get file → | |
| Q5_K_S | 25.1 GB | 26.2 GB | Too big | Get file → | |
| Q5_K_M | 25.9 GB | 27.1 GB | Too big | Get file → | |
| Q5_K_L | 26.2 GB | 27.4 GB | Too big | Get file → | |
| Q6_K | 30.9 GB | 32.1 GB | Too big | Get file → | |
| Q6_K_L | 31.2 GB | 32.4 GB | Too big | Get file → | |
| Q8_0 | 37.8 GB | 39.0 GB | Too big | Get file → | |
| BF16 | 71.1 GB (2 parts) | 72.3 GB | Too big | Get file → |
For 8 GB VRAM, use IQ1_M (9.42 GB file, ~10.6 GB memory). For 12 GB VRAM, IQ2_XS (11.69 GB file, ~12.9 GB memory). For 16 GB VRAM, Q3_K_S (16.4 GB file, ~17.6 GB memory). Larger budgets can use Q4_K_M or Q5_K_M for higher quality.
What GPU do you need for Qwen3.5 35B A3B?
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: the best quant that fits is the main file (0.2 GB file, about 1.4 GB in use).
- 6 GB VRAM: the best quant that fits is the main file (0.2 GB file, about 1.4 GB in use).
- 8 GB VRAM: the best quant that fits is the main file (0.2 GB file, about 1.4 GB in use).
- 12 GB VRAM: the best quant that fits is IQ2_XXS (10.7 GB file, about 11.9 GB in use).
- 16 GB VRAM: the best quant that fits is Q2_K_L (14.0 GB file, about 15.2 GB in use).
- 24 GB VRAM: the best quant that fits is Q4_K_L (22.7 GB file, about 23.9 GB in use).
- 32 GB VRAM: the best quant that fits is Q5_K_L (26.2 GB file, about 27.4 GB in use).
- 48 GB VRAM: the best quant that fits is Q8_0 (37.8 GB file, about 39.0 GB in use).
Which Qwen3.5 35B A3B quant should you choose?
Q4_K_M is the usual starting point for Qwen3.5 35B A3B: it balances file size and quality well. Here it is a 22.3 GB file.
The Q8_0 file is 37.8 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 the first one in the table at 0.2 GB, with the quality trade-offs that come with fewer bits.
New to quant names? Read GGUF quantization explained.
Specifications
Architecture from the GGUF header
Source
Frequently asked questions about Qwen3.5 35B A3B
How much VRAM does Qwen3.5 35B A3B need?
About 23.5 GB at Q4_K_M with an 8K context window. That is the 22.3 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 Qwen3.5 35B A3B on an 8 GB GPU?
Yes, with the smaller files: the 0.2 GB file needs about 1.4 GB at 8K context. Larger quants in this repository do not fit in 8 GB.
What is the file size of Qwen3.5 35B A3B GGUF?
It depends on the quant: file is 0.2 GB, IQ1_M is 9.4 GB, IQ2_XXS is 10.7 GB, IQ2_XS is 11.7 GB, IQ2_S is 11.9 GB, IQ2_M is 13.0 GB and 23 more.
What context length does Qwen3.5 35B A3B support?
The GGUF header lists a trained context length of 262,144 tokens (256K). The KV cache grows with the context you use: about 0.7 GB at 8K and 2.8 GB at 32K.
Where can I download Qwen3.5 35B A3B GGUF?
The files are on Hugging Face in the bartowski/Qwen_Qwen3.5-35B-A3B-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 Qwen3.5 35B A3B 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 Qwen3.5 35B A3B on 8 GB VRAM?
Use the IQ1_M quant, which requires about 10.6 GB of VRAM at 8k context. It will fit on 8 GB with some swapping, but performance may be limited.
Does this model support vision tasks in llama.cpp?
Yes, the model includes vision encoder support, but llama.cpp must be built with the latest imatrix quantization to handle the vision weights.
What license applies to this model?
The model is released under the Apache‑2.0 license, allowing free use and modification.