Qwen3.6-35B-A3B needs about 21.0 GB of memory at its main file (a 19.9 GB file) with an 8K context window, so it fits on a 24 GB GPU. Longer context needs more: at 32K tokens the KV cache alone adds about 2.8 GB.
Qwen3.6-35B-A3B – 35B parameter causal model with vision and Mixture of Experts
Qwen3.6-35B-A3B is a causal language model with a vision encoder. It contains 35B total parameters, 3B active in the MoE. The model supports a 262,144 token context and is quantized to ROCmFP4 4‑bit.
Good at
- Vision‑enabled text generation
- Mixture‑of‑Experts for efficient inference
- Large context handling
Limits
- Requires ROCmFP4 fork for llama.cpp
- No support for -np > 1 or --mmproj with MTP
- License is other, not standard
Best for
- Vision‑based content creation
- Large‑scale inference on AMD GPUs
- Fine‑tuning with Unsloth Studio
Who it’s for. Developers needing a vision‑capable language model with Mixture‑of‑Experts on AMD hardware. Suitable for users who can compile the ROCmFP4 fork and run MTP for faster inference.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| — | 19.9 GB | 21.0 GB | Too big | Get file → |
The single GGUF file is 19.86 GB and needs about 21 GB of VRAM at 8 k context. For 8 GB VRAM use a smaller quant or a different model. For 16 GB VRAM the file fits with some headroom. For 32 GB VRAM the model runs comfortably.
What GPU do you need for Qwen3.6-35B-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: 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 the main file (19.9 GB file, about 21.0 GB in use).
Specifications
Architecture from the GGUF header
Source
Frequently asked questions about Qwen3.6-35B-A3B
How much VRAM does Qwen3.6-35B-A3B need?
About 21.0 GB at its main file with an 8K context window. That is the 19.9 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 Qwen3.6-35B-A3B on an 8 GB GPU?
Not at an 8K context. The smallest file needs about 21.0 GB, so you would need a larger GPU or CPU offloading.
What is the file size of Qwen3.6-35B-A3B GGUF?
It depends on the quant: file is 19.9 GB.
What context length does Qwen3.6-35B-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 Qwen3.6-35B-A3B GGUF?
The files are on Hugging Face in the plunderstruck/Qwen3.6-35B-A3B-MTP-ROCmFP4-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 Qwen3.6-35B-A3B use?
Hugging Face lists the license as other. 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 Qwen3.6-35B-A3B on 8 GB VRAM?
No, the 19.86 GB file requires about 21 GB of VRAM at 8 k context. You would need a smaller quant or more memory.
Does this model support vision?
Yes, the model includes a vision encoder and detected abilities list includes vision.
What backend does it need?
It requires the ROCmFP4 fork and the MTP backend; stock llama.cpp will not load it.