# Qwen3.6 35B-A3B: hardware requirements
> Qwen3.6 35B-A3B (Alibaba Qwen) is a 35B (3B active) mixture-of-experts model with a 256K-token context window, released under the Apache 2.0 license. It needs about 24GB of memory at Q4_K_M with 8K context.
Qwen3.6 35B-A3B is the April 2026 refresh of Qwen's fast mixture-of-experts model, focused on agentic coding and on preserving its reasoning across multi-turn tool use. With only 3B active parameters it generates text very quickly, and its hybrid linear attention keeps the KV cache tiny: the full 262K context needs about 5GB. At Q4_K_M it needs about 24GB with 8K context, which is tight on a 24GB card and comfortable on 32GB GPUs and Macs with 36GB or more; 12GB to 16GB GPUs run it well with experts offloaded to system RAM. It replaces Qwen3 30B-A3B and Qwen3 Coder 30B-A3B for most local users.

## Recommended hardware
- **Minimum GPU:** RTX 4000 Ada (20GB), Q3_K_M, ~88 tokens/sec
- **Best value:** RTX 3090 (24GB), Q3_K_M, ~160 tokens/sec
- **Fastest single GPU:** RTX 5090 (32GB), Q5_K_M, ~195 tokens/sec
- **Mac / unified memory:** Ryzen AI Max+ 395 (128GB), Q8_0, ~30 tokens/sec

## Memory by quantization
| Quant | Weights | Total @ 8K ctx | Total @ 32K ctx |
| --- | --- | --- | --- |
| Q2_K | 13.8GB | 15.0GB | 15.5GB |
| Q3_K_M | 17.9GB | 19.2GB | 19.7GB |
| Q4_K_M | 22.3GB | 23.7GB | 24.2GB |
| Q5_K_M | 26.2GB | 27.7GB | 28.2GB |
| Q6_K | 30.3GB | 32.0GB | 32.5GB |
| Q8_0 | 39.0GB | 41.0GB | 41.5GB |
| BF16 | 73.5GB | 76.5GB | 77.0GB |

## GPUs that run it fully in memory (fastest first)
| GPU | Memory | Quant | Speed | Price |
| --- | --- | --- | --- | --- |
| RTX 5090 | 32GB | Q5_K_M | ~195 tok/s | ~$4,300 |
| RTX 4090 | 24GB | Q3_K_M | ~175 tok/s | ~$2,500 used |
| RTX 3090 | 24GB | Q3_K_M | ~160 tok/s | ~$1,050 used |
| RTX PRO 6000 | 96GB | Q8_0 | ~155 tok/s | ~$16,000 |
| RTX PRO 5000 72GB | 72GB | Q8_0 | ~130 tok/s | ~$9,100 |
| RX 7900 XTX | 24GB | Q3_K_M | ~125 tok/s | ~$1,050 used |
| V100 32GB | 32GB | Q5_K_M | ~105 tok/s | ~$700 used |
| RTX 6000 Ada | 48GB | Q8_0 | ~100 tok/s | ~$4,750 used |
| RTX 4000 Ada | 20GB | Q3_K_M | ~88 tok/s | $1,250 MSRP |
| M5 Ultra 512GB | 512GB | Q8_0 | ~83 tok/s | n/a |
| M5 Ultra 256GB | 256GB | Q8_0 | ~83 tok/s | ~$9,499 |
| RTX A6000 | 48GB | Q8_0 | ~79 tok/s | ~$4,400 used |
| R9700 | 32GB | Q5_K_M | ~78 tok/s | ~$1,800 |
| MI50 32GB | 32GB | Q5_K_M | ~76 tok/s | ~$600 used |
| Arc Pro B60 | 24GB | Q3_K_M | ~66 tok/s | ~$650 |
| M3 Ultra 512GB | 512GB | Q8_0 | ~62 tok/s | $9,499 MSRP |
| Arc Pro B70 | 32GB | Q5_K_M | ~62 tok/s | ~$1,300 |
| Tesla P40 | 24GB | Q3_K_M | ~58 tok/s | ~$230 used |
| M5 Max 128GB | 128GB | Q8_0 | ~56 tok/s | ~$5,100 |
| M4 Max 128GB | 128GB | Q8_0 | ~51 tok/s | $3,499 MSRP |
| DGX Spark | 128GB | Q8_0 | ~32 tok/s | ~$6,950 |
| Ryzen AI Max+ PRO 495 | 192GB | Q8_0 | ~32 tok/s | ~$6,800 |
| Ryzen AI Max+ 395 | 128GB | Q8_0 | ~30 tok/s | ~$3,300 |

## How to run it
- Ollama: `ollama run qwen3.6:35b`
- llama.cpp: `llama-server -hf unsloth/Qwen3.6-35B-A3B-GGUF:Q4_K_M -c 8192 -ngl 99`
- Weights: https://huggingface.co/Qwen/Qwen3.6-35B-A3B

## Strengths
- Very fast: 3B active parameters per token
- Tiny KV cache: 20 KiB per token, about 5GB at the full 262K context
- Strong agentic coding and tool use, with image input
- Apache 2.0 license

## Weaknesses
- The 22GB Q4_K_M file is tight on 24GB GPUs once context is added
- Less world knowledge than dense 27B to 31B models
- Hybrid attention needs recent llama.cpp, MLX, or vLLM builds
Full page: https://llamabuilds.ai/models/qwen3-6-35b-a3b · Hardware guide: https://llamabuilds.ai/models/qwen3-6-35b-a3b/requirements

---
Source: https://llamabuilds.ai/models/qwen3-6-35b-a3b · Updated October 14, 2026 · Llama Builds. Speeds marked ~ are estimates from our bandwidth model (https://llamabuilds.ai/methodology); measured results are labeled as such.