Run unsloth/Qwen3.6-35B-A3B-GGUF locally
unsloth/Qwen3.6-35B-A3B-GGUF is a very large language model with 34.66 billion parameters, built on the qwen35moe architecture. It is released under the apache-2.0 license and has been downloaded 877,585 times.
To run unsloth/Qwen3.6-35B-A3B-GGUF locally at a 4,096-token context, its quantized versions need between 2.11 GB (F16, lowest quality) and 66.73 GB (BF16, highest quality) of memory, weights plus KV cache and a system margin included.
For most users the best balance is F32, needing about 2.94 GB. That means unsloth/Qwen3.6-35B-A3B-GGUF fits entirely in the VRAM of a 6 GB GPU or larger, running fully on the GPU.
All quantizations
| Quant. | Bits | Quality | Weights | KV | Total | Speed~ | Verdict |
|---|---|---|---|---|---|---|---|
| F16 | 0.21 | Very low | 0.84 GB | 0.47 GB | 2.11 GB | 477.6 t/s | Fits in VRAM |
| F32 | 0.41 | Very low | 1.66 GB | 0.47 GB | 2.94 GB | 240.4 t/s | Fits in VRAM |
| IQ1_M | 2.32 | Very low | 9.36 GB | 0.47 GB | 10.63 GB | 5.3 t/s | Offload |
| IQ2_XXS | 2.48 | Very low | 10.02 GB | 0.47 GB | 11.29 GB | 5.0 t/s | Offload |
| IQ2_M | 2.66 | Low | 10.73 GB | 0.47 GB | 12.01 GB | 4.7 t/s | Offload |
| Q2_K_XL | 2.84 | Low | 11.45 GB | 0.47 GB | 12.72 GB | 4.4 t/s | Offload |
| IQ3_XXS | 3.05 | Low | 12.3 GB | 0.47 GB | 13.58 GB | 4.1 t/s | Offload |
| IQ3_S | 3.16 | Low | 12.74 GB | 0.47 GB | 14.01 GB | 3.9 t/s | Offload |
| Q3_K_S | 3.55 | Fair | 14.3 GB | 0.47 GB | 15.58 GB | 3.5 t/s | Offload |
| Q3_K_M | 3.83 | Fair | 15.46 GB | 0.47 GB | 16.74 GB | 3.2 t/s | Offload |
| Q3_K_XL | 3.89 | Fair | 15.69 GB | 0.47 GB | 16.96 GB | 3.2 t/s | Offload |
| IQ4_XS | 4.09 | Fair | 16.51 GB | 0.47 GB | 17.79 GB | 3.0 t/s | Offload |
| IQ4_NL | 4.16 | Fair | 16.8 GB | 0.47 GB | 18.08 GB | 3.0 t/s | Offload |
| IQ4_NL_XL | 4.5 | Good | 18.16 GB | 0.47 GB | 19.44 GB | 2.8 t/s | Offload |
| Q4_K_S | 4.82 | Good | 19.46 GB | 0.47 GB | 20.73 GB | 2.6 t/s | Offload |
| GGUF | 5.01 | Very good | 20.22 GB | 0.47 GB | 21.49 GB | 2.5 t/s | Offload |
| Q4_K_M | 5.11 | Very good | 20.61 GB | 0.47 GB | 21.89 GB | 2.4 t/s | Offload |
| Q4_K_XL | 5.16 | Very good | 20.82 GB | 0.47 GB | 22.1 GB | 2.4 t/s | Offload |
| Q5_K_S | 5.76 | Very good | 23.23 GB | 0.47 GB | 24.5 GB | — | Insufficient |
| Q5_K_M | 6.11 | Very good | 24.64 GB | 0.47 GB | 25.91 GB | — | Insufficient |
| Q5_K_XL | 6.14 | Very good | 24.77 GB | 0.47 GB | 26.04 GB | — | Insufficient |
| Q6_K | 6.76 | Excellent | 27.3 GB | 0.47 GB | 28.57 GB | — | Insufficient |
| Q6_K_XL | 7.35 | Excellent | 29.66 GB | 0.47 GB | 30.93 GB | — | Insufficient |
| Q8_0 | 8.52 | Excellent | 34.37 GB | 0.47 GB | 35.64 GB | — | Insufficient |
| Q8_K_XL | 8.87 | Excellent | 35.81 GB | 0.47 GB | 37.09 GB | — | Insufficient |
| BF16 | 16.22 | Excellent | 65.45 GB | 0.47 GB | 66.73 GB | — | Insufficient |
KV cache computed from the model's exact architecture. Speed is a rough estimate bounded by memory bandwidth.
Frequently asked questions
How much VRAM do you need to run unsloth/Qwen3.6-35B-A3B-GGUF?
You need about 2.94 GB of VRAM to run unsloth/Qwen3.6-35B-A3B-GGUF entirely on the GPU using the F32 quantization (at a 4,096-token context). Smaller quantizations lower the requirement at the cost of quality.
Can I run unsloth/Qwen3.6-35B-A3B-GGUF on an 8 GB GPU?
Yes. With 8 GB of VRAM you can run unsloth/Qwen3.6-35B-A3B-GGUF fully on the GPU using F32 (about 2.94 GB).
Can I run unsloth/Qwen3.6-35B-A3B-GGUF on a 16 GB GPU?
Yes. With 16 GB of VRAM you can run unsloth/Qwen3.6-35B-A3B-GGUF fully on the GPU using Q3_K_S (about 15.58 GB).
Can I run unsloth/Qwen3.6-35B-A3B-GGUF on a 24 GB GPU?
Yes. With 24 GB of VRAM you can run unsloth/Qwen3.6-35B-A3B-GGUF fully on the GPU using Q4_K_XL (about 22.1 GB).
What is the best quantization for unsloth/Qwen3.6-35B-A3B-GGUF?
If memory allows, higher bits-per-weight means better quality. A common sweet spot is a Q4_K_M or Q5_K_M quantization, which keeps most of the quality while roughly halving the memory versus 8-bit. Pick the highest quantization that still fits in your VRAM.