← CanIRunLocal · All hardware · All models
Alibaba · 36B parameters (3B active) · coding, chat, reasoning, vision.
Ollama weight files: Q4_K_M 24 GB (qwen3.6:35b-a3b-q4_K_M); Q8_0 39 GB (qwen3.6:35b-a3b-q8_0).
Yes on 44 machines in this list. The smallest is RTX 4060 Ti 16GB: Q4_K_M, about 24 GB, roughly 22 tokens per second, via LM Studio. lms get unsloth/Qwen3.6-35B-A3B-GGUF@Q4_K_M
ollama run qwen3.6:35b-a3b-q8_0ollama run qwen3.6:35b-a3b-q8_0ollama run qwen3.6:35b-a3b-q8_0ollama run qwen3.6:35b-a3b-q8_0lms get unsloth/Qwen3.6-35B-A3B-GGUF@Q8_0llama-server -hf unsloth/Qwen3.6-35B-A3B-GGUF:Q8_0 -c 8192vllm serve Qwen/Qwen3.6-35B-A3B --max-model-len 8192 --gpu-memory-utilization 0.9| Hardware | Runtime | Quant | Memory | Fit | Speed | Command |
|---|---|---|---|---|---|---|
| RTX 4060 Ti 16GB | LM Studio | Q4_K_M GGUF | 24 GB | partial offload (slower) | ~22 tok/s | lms get unsloth/Qwen3.6-35B-A3B-GGUF@Q4_K_M |
| RTX 4070 Ti Super 16GB | LM Studio | Q4_K_M GGUF | 24 GB | partial offload (slower) | ~26 tok/s | lms get unsloth/Qwen3.6-35B-A3B-GGUF@Q4_K_M |
| RTX 4080 16GB | LM Studio | Q4_K_M GGUF | 24 GB | partial offload (slower) | ~26 tok/s | lms get unsloth/Qwen3.6-35B-A3B-GGUF@Q4_K_M |
| RTX 5080 16GB | LM Studio | Q4_K_M GGUF | 24 GB | partial offload (slower) | ~27 tok/s | lms get unsloth/Qwen3.6-35B-A3B-GGUF@Q4_K_M |
| RX 7800 XT 16GB | LM Studio | Q4_K_M GGUF | 24 GB | partial offload (slower) | ~26 tok/s | lms get unsloth/Qwen3.6-35B-A3B-GGUF@Q4_K_M |
| RX 9070 XT 16GB | LM Studio | Q4_K_M GGUF | 24 GB | partial offload (slower) | ~26 tok/s | lms get unsloth/Qwen3.6-35B-A3B-GGUF@Q4_K_M |
| RTX 3090 24GB | Ollama | Q4_K_M | 25.9 GB | partial offload (slower) | ~64 tok/s | ollama run qwen3.6:35b-a3b-q4_K_M |
| RTX 4090 24GB | Ollama | Q4_K_M | 25.9 GB | partial offload (slower) | ~65 tok/s | ollama run qwen3.6:35b-a3b-q4_K_M |
| RX 7900 XTX 24GB | Ollama | Q4_K_M | 25.9 GB | partial offload (slower) | ~64 tok/s | ollama run qwen3.6:35b-a3b-q4_K_M |
| M1 Pro 32GB | LM Studio | MLX 4bit | 22.3 GB | fits in fast memory | ~47 tok/s | lms get mlx-community/Qwen3.6-35B-A3B-4bit |
| M2 Pro 32GB | LM Studio | MLX 4bit | 22.3 GB | fits in fast memory | ~47 tok/s | lms get mlx-community/Qwen3.6-35B-A3B-4bit |
| M4 32GB | LM Studio | MLX 4bit | 22.3 GB | fits in fast memory | ~28 tok/s | lms get mlx-community/Qwen3.6-35B-A3B-4bit |
| M3 Pro 36GB | Ollama | Q4_K_M | 25.9 GB | fits in fast memory | ~30 tok/s | ollama run qwen3.6:35b-a3b-q4_K_M |
| Mac Studio M4 Max (32c GPU) 36GB | Ollama | Q4_K_M | 25.9 GB | fits in fast memory | ~82 tok/s | ollama run qwen3.6:35b-a3b-q4_K_M |
| Mac Studio M5 Max (32c GPU) 36GB | Ollama | Q4_K_M | 25.9 GB | fits in fast memory | ~92 tok/s | ollama run qwen3.6:35b-a3b-q4_K_M |
| RTX 5090 32GB | Ollama | Q4_K_M | 25.9 GB | fits in fast memory | ~269 tok/s | ollama run qwen3.6:35b-a3b-q4_K_M |
| M4 Pro 48GB | Ollama | Q4_K_M | 25.9 GB | fits in fast memory | ~55 tok/s | ollama run qwen3.6:35b-a3b-q4_K_M |
| Mac Studio M4 Max (40c GPU) 48GB | Ollama | Q4_K_M | 25.9 GB | fits in fast memory | ~109 tok/s | ollama run qwen3.6:35b-a3b-q4_K_M |
| Mac Studio M5 Max (40c GPU) 48GB | Ollama | Q4_K_M | 25.9 GB | fits in fast memory | ~123 tok/s | ollama run qwen3.6:35b-a3b-q4_K_M |
| M1 Max 64GB | Ollama | Q8_0 | 40.9 GB | fits in fast memory | ~49 tok/s | ollama run qwen3.6:35b-a3b-q8_0 |
| M2 Max 64GB | Ollama | Q8_0 | 40.9 GB | fits in fast memory | ~49 tok/s | ollama run qwen3.6:35b-a3b-q8_0 |
| M3 Max 64GB | Ollama | Q8_0 | 40.9 GB | fits in fast memory | ~49 tok/s | ollama run qwen3.6:35b-a3b-q8_0 |
| M4 Max 64GB | Ollama | Q8_0 | 40.9 GB | fits in fast memory | ~67 tok/s | ollama run qwen3.6:35b-a3b-q8_0 |
| Mac Studio M4 Max (40c GPU) 64GB | Ollama | Q8_0 | 40.9 GB | fits in fast memory | ~67 tok/s | ollama run qwen3.6:35b-a3b-q8_0 |
| Mac Studio M5 Max (40c GPU) 64GB | Ollama | Q8_0 | 40.9 GB | fits in fast memory | ~76 tok/s | ollama run qwen3.6:35b-a3b-q8_0 |
| DGX Spark 64GB (ASUS, Dell, HP, MSI, Acer, Gigabyte) | Ollama | Q8_0 | 40.9 GB | fits in fast memory | ~48 tok/s | ollama run qwen3.6:35b-a3b-q8_0 |
| Mac Studio M3 Ultra 96GB | Ollama | Q8_0 | 40.9 GB | fits in fast memory | ~101 tok/s | ollama run qwen3.6:35b-a3b-q8_0 |
| Mac Studio M5 Ultra 96GB | Ollama | Q8_0 | 40.9 GB | fits in fast memory | ~148 tok/s | ollama run qwen3.6:35b-a3b-q8_0 |
| M3 Max 128GB | Ollama | Q8_0 | 40.9 GB | fits in fast memory | ~49 tok/s | ollama run qwen3.6:35b-a3b-q8_0 |
| M4 Max 128GB | Ollama | Q8_0 | 40.9 GB | fits in fast memory | ~67 tok/s | ollama run qwen3.6:35b-a3b-q8_0 |
| Mac Studio M4 Max (40c GPU) 128GB | Ollama | Q8_0 | 40.9 GB | fits in fast memory | ~67 tok/s | ollama run qwen3.6:35b-a3b-q8_0 |
| Mac Studio M5 Max (40c GPU) 128GB | Ollama | Q8_0 | 40.9 GB | fits in fast memory | ~76 tok/s | ollama run qwen3.6:35b-a3b-q8_0 |
| DGX Spark 128GB (NVIDIA, ASUS, Dell, Lenovo, HP, MSI, Acer, Gigabyte) | Ollama | Q8_0 | 40.9 GB | fits in fast memory | ~48 tok/s | ollama run qwen3.6:35b-a3b-q8_0 |
| M2 Ultra 192GB | Ollama | Q8_0 | 40.9 GB | fits in fast memory | ~98 tok/s | ollama run qwen3.6:35b-a3b-q8_0 |
| 2x Mac Studio M5 Max (40c GPU) 128GB | llama.cpp | Q8_0 GGUF | 38.8 GB | fits in fast memory | ~68 tok/s | llama-server -hf unsloth/Qwen3.6-35B-A3B-GGUF:Q8_0 --rpc node2:50052 -c 8192 # run rpc-server on each other node |
| Mac Studio M3 Ultra 256GB | Ollama | Q8_0 | 40.9 GB | fits in fast memory | ~101 tok/s | ollama run qwen3.6:35b-a3b-q8_0 |
| Mac Studio M5 Ultra (64c GPU) 256GB | Ollama | Q8_0 | 40.9 GB | fits in fast memory | ~148 tok/s | ollama run qwen3.6:35b-a3b-q8_0 |
| Mac Studio M5 Ultra (80c GPU) 256GB | Ollama | Q8_0 | 40.9 GB | fits in fast memory | ~148 tok/s | ollama run qwen3.6:35b-a3b-q8_0 |
| 2x DGX Spark 128GB | llama.cpp | Q8_0 GGUF | 38.8 GB | fits in fast memory | ~43 tok/s | llama-server -hf unsloth/Qwen3.6-35B-A3B-GGUF:Q8_0 --rpc node2:50052 -c 8192 # run rpc-server on each other node |
| 4x Mac Studio M5 Max (40c GPU) 128GB | llama.cpp | Q8_0 GGUF | 38.8 GB | fits in fast memory | ~60 tok/s | llama-server -hf unsloth/Qwen3.6-35B-A3B-GGUF:Q8_0 --rpc node2:50052,node3:50052,node4:50052 -c 8192 # run rpc-server on each other node |
| 2x Mac Studio M5 Ultra (80c GPU) 256GB | llama.cpp | Q8_0 GGUF | 38.8 GB | fits in fast memory | ~133 tok/s | llama-server -hf unsloth/Qwen3.6-35B-A3B-GGUF:Q8_0 --rpc node2:50052 -c 8192 # run rpc-server on each other node |
| Mac Studio M3 Ultra (80c GPU) 512GB | Ollama | Q8_0 | 40.9 GB | fits in fast memory | ~101 tok/s | ollama run qwen3.6:35b-a3b-q8_0 |
| 4x DGX Spark 128GB | llama.cpp | Q8_0 GGUF | 38.8 GB | fits in fast memory | ~38 tok/s | llama-server -hf unsloth/Qwen3.6-35B-A3B-GGUF:Q8_0 --rpc node2:50052,node3:50052,node4:50052 -c 8192 # run rpc-server on each other node |
| 4x Mac Studio M5 Ultra (80c GPU) 256GB | llama.cpp | Q8_0 GGUF | 38.8 GB | fits in fast memory | ~117 tok/s | llama-server -hf unsloth/Qwen3.6-35B-A3B-GGUF:Q8_0 --rpc node2:50052,node3:50052,node4:50052 -c 8192 # run rpc-server on each other node |
Estimates, not benchmarks. Model sizes are ollama.com download sizes; memory needed adds headroom for ~8k tokens of context; speeds are rough figures from memory bandwidth. Data updated 2026-10-18.
A project by OnDevice Labs · llms.txt