Models / Qwen3-VL
Qwen3-VL by Qwen
10 models (2B to 235B) · ↓ 15.8M downloads a month · from $0.10/h on RTX 3060 right now. Each model's versions do the same job; quantized ones (AWQ, FP8, 4-bit) need less GPU memory and run on cheaper machines.
Qwen3-VL 2B
| Version | Runtime | Weights | GPU memory | Cheapest machine now |
|---|
| OriginalQwen/Qwen3-VL-2B-Instruct | vLLM | 4.3 GB | 8 GB | $0.10/h RTX 3060 |
Qwen3-VL 2B
| Version | Runtime | Weights | GPU memory | Cheapest machine now |
|---|
| OriginalQwen/Qwen3-VL-2B-Thinking | vLLM | 4.3 GB | 8 GB | $0.10/h RTX 3060 |
Qwen3-VL 4B
| Version | Runtime | Weights | GPU memory | Cheapest machine now |
|---|
| OriginalQwen/Qwen3-VL-4B-Instruct | vLLM | 8.9 GB | 13.4 GB | $0.14/h RTX A4000 |
| FP8Qwen/Qwen3-VL-4B-Instruct-FP8 | vLLM | 6.0 GB | 10.3 GB | $0.10/h RTX 3060 |
Qwen3-VL 8B
| Version | Runtime | Weights | GPU memory | Cheapest machine now |
|---|
| OriginalQwen/Qwen3-VL-8B-Instruct | vLLM | 18 GB | 23 GB | $0.24/h RTX 3090 |
| FP8Qwen/Qwen3-VL-8B-Instruct-FP8 | vLLM | 11 GB | 15.3 GB | $0.14/h RTX A4000 |
| AWQ 4bit (cyankiwi)cyankiwi/Qwen3-VL-8B-Instruct-AWQ-4bit | vLLM | 7.5 GB | 12 GB | $0.10/h RTX 3060 |
| quantized.w4a16 (SherlockID365)SherlockID365/Qwen3-VL-8B-Instruct-quantized.w4a16 | vLLM | 7.2 GB | 11.6 GB | $0.10/h RTX 3060 |
Qwen3-VL 8B
| Version | Runtime | Weights | GPU memory | Cheapest machine now |
|---|
| OriginalQwen/Qwen3-VL-8B-Thinking | vLLM | 18 GB | 23 GB | $0.24/h RTX 3090 |
Qwen3-VL 30B
| Version | Runtime | Weights | GPU memory | Cheapest machine now |
|---|
| OriginalQwen/Qwen3-VL-30B-A3B-Instruct | vLLM | 62 GB | 72.2 GB | $0.89/h 2x Nvidia A40 |
| FP8Qwen/Qwen3-VL-30B-A3B-Instruct-FP8 | vLLM | 32 GB | 39 GB | $0.45/h Nvidia A40 |
| AWQ (QuantTrio)QuantTrio/Qwen3-VL-30B-A3B-Instruct-AWQ | vLLM | 18 GB | 23 GB | $0.24/h RTX 3090 |
| AWQ 4bit (cyankiwi)cyankiwi/Qwen3-VL-30B-A3B-Instruct-AWQ-4bit | vLLM | 19 GB | 24.4 GB | $0.34/h 2x RTX 5060 Ti |
Qwen3-VL 30B
| Version | Runtime | Weights | GPU memory | Cheapest machine now |
|---|
| AWQ (QuantTrio)QuantTrio/Qwen3-VL-30B-A3B-Thinking-AWQ | vLLM | 18 GB | 23 GB | $0.24/h RTX 3090 |
Qwen3-VL 32B
| Version | Runtime | Weights | GPU memory | Cheapest machine now |
|---|
| OriginalQwen/Qwen3-VL-32B-Instruct | vLLM | 67 GB | 78.7 GB | $0.89/h 2x Nvidia A40 |
| FP8Qwen/Qwen3-VL-32B-Instruct-FP8 | vLLM | 36 GB | 44.1 GB | $0.45/h Nvidia A40 |
| AWQ (QuantTrio)QuantTrio/Qwen3-VL-32B-Instruct-AWQ | vLLM | 21 GB | 27.4 GB | $0.34/h 2x RTX 5060 Ti |
| NVFP4 (RedHatAI)RedHatAI/Qwen3-VL-32B-Instruct-NVFP4 | vLLM | 22 GB | 28.9 GB | $0.34/h 2x RTX 5060 Ti |
Qwen3-VL 235B
| Version | Runtime | Weights | GPU memory | Cheapest machine now |
|---|
| OriginalQwen/Qwen3-VL-235B-A22B-Instruct | vLLM | 471 GB | 527.7 GB | $11.77/h 8x RTX PRO 6000 |
| FP8Qwen/Qwen3-VL-235B-A22B-Instruct-FP8 | vLLM | 238 GB | 268 GB | $5.33/h 8x Nvidia A100 |
| NVFP4 (RedHatAI)RedHatAI/Qwen3-VL-235B-A22B-Instruct-NVFP4 | vLLM | 135 GB | 154.3 GB | $1.38/h 8x RTX 3090 |
Qwen3-VL 235B
| Version | Runtime | Weights | GPU memory | Cheapest machine now |
|---|
| OriginalQwen/Qwen3-VL-235B-A22B-Thinking | vLLM | 471 GB | 527.7 GB | $11.77/h 8x RTX PRO 6000 |
Model details come from Hugging Face and are refreshed daily; prices follow the live offers. GPU memory is GridPup's estimate: the weights, room for an 8,192-token conversation (chat models) and the runtime's own memory. Large models are split across the GPUs of one machine ("4x RTX 4090").