← CanIRunLocal · All hardware · All models
Alibaba · 4.44B parameters · vision, chat, reasoning.
Ollama weight files: Q4_K_M 3.3 GB (qwen3-vl:4b); Q8_0 5.1 GB (qwen3-vl:4b-instruct-q8_0).
Yes on 62 machines in this list. The smallest is M1 8GB: Q4_K_M, about 4.5 GB, roughly 16 tokens per second, via Ollama. ollama run qwen3-vl:4b
ollama run qwen3-vl:4b-instruct-q8_0ollama run qwen3-vl:4b-instruct-q8_0ollama run qwen3-vl:4b-instruct-q8_0ollama run qwen3-vl:4b-instruct-q8_0lms get unsloth/Qwen3-VL-4B-Instruct-GGUF@Q8_0llama-server -hf unsloth/Qwen3-VL-4B-Instruct-GGUF:Q8_0 -c 8192vllm serve Qwen/Qwen3-VL-4B-Instruct --max-model-len 8192 --gpu-memory-utilization 0.9| Hardware | Runtime | Quant | Memory | Fit | Speed | Command |
|---|---|---|---|---|---|---|
| M1 8GB | Ollama | Q4_K_M | 4.5 GB | fits in fast memory | ~16 tok/s | ollama run qwen3-vl:4b |
| M2 8GB | Ollama | Q4_K_M | 4.5 GB | fits in fast memory | ~24 tok/s | ollama run qwen3-vl:4b |
| M3 8GB | Ollama | Q4_K_M | 4.5 GB | fits in fast memory | ~24 tok/s | ollama run qwen3-vl:4b |
| RTX 3060 Ti 8GB | Ollama | Q8_0 | 6.3 GB | fits in fast memory | ~53 tok/s | ollama run qwen3-vl:4b-instruct-q8_0 |
| RTX 3070 8GB | Ollama | Q8_0 | 6.3 GB | fits in fast memory | ~53 tok/s | ollama run qwen3-vl:4b-instruct-q8_0 |
| RTX 4060 8GB | Ollama | Q8_0 | 6.3 GB | fits in fast memory | ~32 tok/s | ollama run qwen3-vl:4b-instruct-q8_0 |
| RX 7600 8GB | Ollama | Q8_0 | 6.3 GB | fits in fast memory | ~34 tok/s | ollama run qwen3-vl:4b-instruct-q8_0 |
| RTX 3080 10GB | Ollama | Q8_0 | 6.3 GB | fits in fast memory | ~89 tok/s | ollama run qwen3-vl:4b-instruct-q8_0 |
| M1 16GB | Ollama | Q8_0 | 6.3 GB | fits in fast memory | ~11 tok/s | ollama run qwen3-vl:4b-instruct-q8_0 |
| M2 16GB | Ollama | Q8_0 | 6.3 GB | fits in fast memory | ~16 tok/s | ollama run qwen3-vl:4b-instruct-q8_0 |
| M3 16GB | Ollama | Q8_0 | 6.3 GB | fits in fast memory | ~16 tok/s | ollama run qwen3-vl:4b-instruct-q8_0 |
| M4 16GB | Ollama | Q8_0 | 6.3 GB | fits in fast memory | ~19 tok/s | ollama run qwen3-vl:4b-instruct-q8_0 |
| RTX 3060 12GB | Ollama | Q8_0 | 6.3 GB | fits in fast memory | ~42 tok/s | ollama run qwen3-vl:4b-instruct-q8_0 |
| RTX 4070 12GB | Ollama | Q8_0 | 6.3 GB | fits in fast memory | ~59 tok/s | ollama run qwen3-vl:4b-instruct-q8_0 |
| RTX 5070 12GB | Ollama | Q8_0 | 6.3 GB | fits in fast memory | ~79 tok/s | ollama run qwen3-vl:4b-instruct-q8_0 |
| RTX 4060 Ti 16GB | Ollama | Q8_0 | 6.3 GB | fits in fast memory | ~34 tok/s | ollama run qwen3-vl:4b-instruct-q8_0 |
| RTX 4070 Ti Super 16GB | Ollama | Q8_0 | 6.3 GB | fits in fast memory | ~79 tok/s | ollama run qwen3-vl:4b-instruct-q8_0 |
| RTX 4080 16GB | Ollama | Q8_0 | 6.3 GB | fits in fast memory | ~84 tok/s | ollama run qwen3-vl:4b-instruct-q8_0 |
| RTX 5080 16GB | Ollama | Q8_0 | 6.3 GB | fits in fast memory | ~113 tok/s | ollama run qwen3-vl:4b-instruct-q8_0 |
| RX 7800 XT 16GB | Ollama | Q8_0 | 6.3 GB | fits in fast memory | ~73 tok/s | ollama run qwen3-vl:4b-instruct-q8_0 |
| RX 9070 XT 16GB | Ollama | Q8_0 | 6.3 GB | fits in fast memory | ~75 tok/s | ollama run qwen3-vl:4b-instruct-q8_0 |
| M2 24GB | Ollama | Q8_0 | 6.3 GB | fits in fast memory | ~16 tok/s | ollama run qwen3-vl:4b-instruct-q8_0 |
| M3 24GB | Ollama | Q8_0 | 6.3 GB | fits in fast memory | ~16 tok/s | ollama run qwen3-vl:4b-instruct-q8_0 |
| M4 24GB | Ollama | Q8_0 | 6.3 GB | fits in fast memory | ~19 tok/s | ollama run qwen3-vl:4b-instruct-q8_0 |
| RTX 3090 24GB | Ollama | Q8_0 | 6.3 GB | fits in fast memory | ~110 tok/s | ollama run qwen3-vl:4b-instruct-q8_0 |
| RTX 4090 24GB | Ollama | Q8_0 | 6.3 GB | fits in fast memory | ~119 tok/s | ollama run qwen3-vl:4b-instruct-q8_0 |
| RX 7900 XTX 24GB | Ollama | Q8_0 | 6.3 GB | fits in fast memory | ~113 tok/s | ollama run qwen3-vl:4b-instruct-q8_0 |
| M1 Pro 32GB | Ollama | Q8_0 | 6.3 GB | fits in fast memory | ~31 tok/s | ollama run qwen3-vl:4b-instruct-q8_0 |
| M2 Pro 32GB | Ollama | Q8_0 | 6.3 GB | fits in fast memory | ~31 tok/s | ollama run qwen3-vl:4b-instruct-q8_0 |
| M4 32GB | Ollama | Q8_0 | 6.3 GB | fits in fast memory | ~19 tok/s | ollama run qwen3-vl:4b-instruct-q8_0 |
| M3 Pro 36GB | Ollama | Q8_0 | 6.3 GB | fits in fast memory | ~24 tok/s | ollama run qwen3-vl:4b-instruct-q8_0 |
| Mac Studio M4 Max (32c GPU) 36GB | Ollama | Q8_0 | 6.3 GB | fits in fast memory | ~64 tok/s | ollama run qwen3-vl:4b-instruct-q8_0 |
| Mac Studio M5 Max (32c GPU) 36GB | Ollama | Q8_0 | 6.3 GB | fits in fast memory | ~72 tok/s | ollama run qwen3-vl:4b-instruct-q8_0 |
| RTX 5090 32GB | Ollama | Q8_0 | 6.3 GB | fits in fast memory | ~211 tok/s | ollama run qwen3-vl:4b-instruct-q8_0 |
| M4 Pro 48GB | Ollama | Q8_0 | 6.3 GB | fits in fast memory | ~43 tok/s | ollama run qwen3-vl:4b-instruct-q8_0 |
| Mac Studio M4 Max (40c GPU) 48GB | Ollama | Q8_0 | 6.3 GB | fits in fast memory | ~86 tok/s | ollama run qwen3-vl:4b-instruct-q8_0 |
| Mac Studio M5 Max (40c GPU) 48GB | Ollama | Q8_0 | 6.3 GB | fits in fast memory | ~96 tok/s | ollama run qwen3-vl:4b-instruct-q8_0 |
| M1 Max 64GB | Ollama | Q8_0 | 6.3 GB | fits in fast memory | ~63 tok/s | ollama run qwen3-vl:4b-instruct-q8_0 |
| M2 Max 64GB | Ollama | Q8_0 | 6.3 GB | fits in fast memory | ~63 tok/s | ollama run qwen3-vl:4b-instruct-q8_0 |
| M3 Max 64GB | Ollama | Q8_0 | 6.3 GB | fits in fast memory | ~63 tok/s | ollama run qwen3-vl:4b-instruct-q8_0 |
| M4 Max 64GB | Ollama | Q8_0 | 6.3 GB | fits in fast memory | ~86 tok/s | ollama run qwen3-vl:4b-instruct-q8_0 |
| Mac Studio M4 Max (40c GPU) 64GB | Ollama | Q8_0 | 6.3 GB | fits in fast memory | ~86 tok/s | ollama run qwen3-vl:4b-instruct-q8_0 |
| Mac Studio M5 Max (40c GPU) 64GB | Ollama | Q8_0 | 6.3 GB | fits in fast memory | ~96 tok/s | ollama run qwen3-vl:4b-instruct-q8_0 |
| DGX Spark 64GB (ASUS, Dell, HP, MSI, Acer, Gigabyte) | Ollama | Q8_0 | 6.3 GB | fits in fast memory | ~40 tok/s | ollama run qwen3-vl:4b-instruct-q8_0 |
| Mac Studio M3 Ultra 96GB | Ollama | Q8_0 | 6.3 GB | fits in fast memory | ~128 tok/s | ollama run qwen3-vl:4b-instruct-q8_0 |
| Mac Studio M5 Ultra 96GB | Ollama | Q8_0 | 6.3 GB | fits in fast memory | ~188 tok/s | ollama run qwen3-vl:4b-instruct-q8_0 |
| M3 Max 128GB | Ollama | Q8_0 | 6.3 GB | fits in fast memory | ~63 tok/s | ollama run qwen3-vl:4b-instruct-q8_0 |
| M4 Max 128GB | Ollama | Q8_0 | 6.3 GB | fits in fast memory | ~86 tok/s | ollama run qwen3-vl:4b-instruct-q8_0 |
| Mac Studio M4 Max (40c GPU) 128GB | Ollama | Q8_0 | 6.3 GB | fits in fast memory | ~86 tok/s | ollama run qwen3-vl:4b-instruct-q8_0 |
| Mac Studio M5 Max (40c GPU) 128GB | Ollama | Q8_0 | 6.3 GB | fits in fast memory | ~96 tok/s | ollama run qwen3-vl:4b-instruct-q8_0 |
| DGX Spark 128GB (NVIDIA, ASUS, Dell, Lenovo, HP, MSI, Acer, Gigabyte) | Ollama | Q8_0 | 6.3 GB | fits in fast memory | ~40 tok/s | ollama run qwen3-vl:4b-instruct-q8_0 |
| M2 Ultra 192GB | Ollama | Q8_0 | 6.3 GB | fits in fast memory | ~125 tok/s | ollama run qwen3-vl:4b-instruct-q8_0 |
| 2x Mac Studio M5 Max (40c GPU) 128GB | llama.cpp | Q8_0 GGUF | 5.5 GB | fits in fast memory | ~98 tok/s | llama-server -hf unsloth/Qwen3-VL-4B-Instruct-GGUF:Q8_0 --rpc node2:50052 -c 8192 # run rpc-server on each other node |
| Mac Studio M3 Ultra 256GB | Ollama | Q8_0 | 6.3 GB | fits in fast memory | ~128 tok/s | ollama run qwen3-vl:4b-instruct-q8_0 |
| Mac Studio M5 Ultra (64c GPU) 256GB | Ollama | Q8_0 | 6.3 GB | fits in fast memory | ~188 tok/s | ollama run qwen3-vl:4b-instruct-q8_0 |
| Mac Studio M5 Ultra (80c GPU) 256GB | Ollama | Q8_0 | 6.3 GB | fits in fast memory | ~188 tok/s | ollama run qwen3-vl:4b-instruct-q8_0 |
| 2x DGX Spark 128GB | llama.cpp | Q8_0 GGUF | 5.5 GB | fits in fast memory | ~41 tok/s | llama-server -hf unsloth/Qwen3-VL-4B-Instruct-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 | 5.5 GB | fits in fast memory | ~86 tok/s | llama-server -hf unsloth/Qwen3-VL-4B-Instruct-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 | 5.5 GB | fits in fast memory | ~191 tok/s | llama-server -hf unsloth/Qwen3-VL-4B-Instruct-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 | 6.3 GB | fits in fast memory | ~128 tok/s | ollama run qwen3-vl:4b-instruct-q8_0 |
| 4x DGX Spark 128GB | llama.cpp | Q8_0 GGUF | 5.5 GB | fits in fast memory | ~36 tok/s | llama-server -hf unsloth/Qwen3-VL-4B-Instruct-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 | 5.5 GB | fits in fast memory | ~168 tok/s | llama-server -hf unsloth/Qwen3-VL-4B-Instruct-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