Qwen3.6 35B-A3B: VRAM requirements & speed
Qwen3.6 35B-A3B (35.95B MoE, 3B active, apache-2.0). Native context 256k, 1.01M with YaRN. Smallest GGUF: 22.13 GB.
Memory needed by quantization and context
GGUF file size plus f16 KV cache (and fixed overhead of ~1 GB per GPU not shown). Source: unsloth/Qwen3.6-35B-A3B-GGUF.
| Quant | File | +8k ctx | +32k ctx | +128k ctx | +256k ctx | +1M ctx |
|---|---|---|---|---|---|---|
| Q4_K_M UD-Q4_K_M | 22.13 GB | 22.4 GB | 22.9 GB | 24.9 GB | 27.6 GB | 42.7 GB |
| Q5_K_M UD-Q5_K_M | 26.46 GB | 26.7 GB | 27.2 GB | 29.2 GB | 31.9 GB | 47.0 GB |
| Q6_K UD-Q6_K | 29.31 GB | 29.5 GB | 30.0 GB | 32.1 GB | 34.7 GB | 49.9 GB |
| Q8_0 | 36.9 GB | 37.1 GB | 37.6 GB | 39.7 GB | 42.3 GB | 57.4 GB |
KV cache size by context length
Computed from the model's attention layout, including a fixed 0.07 GB linear-attention state.
| Context | f16 | q8_0 | q4_0 |
|---|---|---|---|
| 8k | 0.23 GB | 0.16 GB | 0.11 GB |
| 32k | 0.74 GB | 0.42 GB | 0.25 GB |
| 128k | 2.75 GB | 1.49 GB | 0.82 GB |
| 256k | 5.43 GB | 2.92 GB | 1.58 GB |
| 1M YaRN | 20.55 GB | 10.95 GB | 5.83 GB |
Which hardware runs Qwen3.6 35B-A3B? · 16k context
| Hardware | Quant | Fit | Memory | Tok/s | Speed source |
|---|---|---|---|---|---|
| RTX PRO 6000 96GB (bonus page) | Q8_0 | fits fully | 37.3 GB | 281 | estimated |
| RTX 5090 32GB | UD-Q6_K | fits fully | 29.7 GB | 260 | measured |
| AMD RX 7900 XTX 24GB | UD-Q4_K_M | fits fully | 22.5 GB | 248 | estimated |
| Mac Studio M5 Ultra 256GB | Q8_0 | fits fully | 37.3 GB | 193 | estimated |
| Mac Studio M5 Ultra 512GB (ships late Oct 2026) | Q8_0 | fits fully | 37.3 GB | 193 | estimated |
| RTX 4090 24GB | UD-Q4_K_M | fits fully | 22.5 GB | 190 | measured |
| 2x RTX 3090 (48GB total) | Q8_0 | fits fully | 37.3 GB | 163 | estimated |
| RTX 3090 24GB | UD-Q4_K_M | fits fully | 22.5 GB | 159 | measured |
| Mac Studio (2025) M3 Ultra 96GB | Q8_0 | fits fully | 37.3 GB | 83 | estimated |
| RTX 5080 16GB | UD-Q4_K_M | MoE offload to CPU | 22.5 GB | 72 | estimated |
| Mac Studio M5 Max 128GB | Q8_0 | fits fully | 37.3 GB | 71 | measured |
| MacBook Pro 14/16 M5 Max 128GB | Q8_0 | fits fully | 37.3 GB | 71 | measured |
| RTX 5070 Ti 16GB | UD-Q4_K_M | MoE offload to CPU | 22.5 GB | 71 | estimated |
| MacBook Pro 14/16 M5 Pro 48GB | UD-Q6_K | fits fully | 29.7 GB | 68 | estimated |
| MacBook Pro 14/16 (2024) M4 Max 128GB | Q8_0 | fits fully | 37.3 GB | 66 | estimated |
| 16GB VRAM class (RTX 4060 Ti 16GB, 32GB RAM) | UD-Q4_K_M | MoE offload to CPU | 22.5 GB | 60 | estimated |
| AMD Ryzen AI Max+ 395 (Strix Halo) 128GB | Q8_0 | fits fully | 37.3 GB | 53 | measured |
| RTX 5070 12GB | UD-Q4_K_M | MoE offload to CPU | 22.5 GB | 50 | estimated |
| NVIDIA DGX Spark 128GB | Q8_0 | fits fully | 37.3 GB | 49 | measured |
| RTX 4070 12GB | UD-Q4_K_M | MoE offload to CPU | 22.5 GB | 49 | estimated |
| Mac mini (2024) M4 Pro 64GB | Q8_0 | fits fully | 37.3 GB | 40 | estimated |
| 8GB VRAM class (RTX 4060 8GB, 32GB RAM) | UD-Q4_K_M | MoE offload to CPU | 22.5 GB | 36 | estimated |
| RTX 3060 12GB (32GB RAM) | UD-Q4_K_M | MoE offload to CPU | 22.5 GB | 28 | estimated |
| Mac mini / iMac M4 16GB | - | does not fit | - | - | |
| Mac mini M6 32GB | - | does not fit | - | - | |
| MacBook Air M5 32GB | - | does not fit | - | - |
Discrete GPUs are modelled with 32 GB of system RAM available for offload. Does not fit even with offload: Mac mini / iMac M4 16GB, Mac mini M6 32GB, MacBook Air M5 32GB.
Links
Model card · GGUF files · Open in the advisor · last verified 2026-10-14
FAQ
How much VRAM does Qwen3.6 35B-A3B need?
The smallest quant (Q4_K_M) is 22.13 GB, so you need roughly 23 GB for an 8k context including KV cache and overhead. At 1M context the f16 KV cache alone is 20.5 GB.
Can I run Qwen3.6 35B-A3B on a RTX PRO 6000 96GB (bonus page)?
Yes: on RTX PRO 6000 96GB (bonus page) it fits fully as Q8_0 at about 281 tok/s (estimated). 16 of 26 hardware configurations fit it fully at 16k context.
What is the maximum context for Qwen3.6 35B-A3B?
256k tokens natively, up to 1.01M with YaRN as documented on the model card. Memory, not the model, is usually the limit: see the KV cache table above.
All models: model index · by memory: 8 GB · 12 GB · 16 GB · 24 GB · 32 GB · 48 GB · 64 GB · 96 GB · 128 GB · 192 GB · 256 GB · 512 GB. Numbers use the same fit math as the advisor; sources and methods.