Qwen3-VL-8B-Instruct
by Qwen
8.8B params · image-text-to-text · 808 likes · 8.1M downloads
Qwen3-VL-8B-Instruct is a 8.8B parameter model. At Q4 quantization it requires 4GB of VRAM. It runs comfortably on GeForce RTX 4090 (149 tok/s), GeForce RTX 5090 (224 tok/s), M4 Max 128GB (54 tok/s).
Inference providers
| Provider | $/1M in | $/1M out | Throughput |
|---|---|---|---|
| Novita | 67 tok/s | ||
| Together AI | 63 tok/s |
GPU compatibility
| GPU | VRAM | Q4 Decode | Verdict |
|---|---|---|---|
| GeForce RTX 4090 | 24GB | 149 tok/s | comfortable |
| GeForce RTX 5090 | 32GB | 224 tok/s | comfortable |
| M4 Max 128GB | 128GB | 54 tok/s | comfortable |
| M4 Pro 48GB | 48GB | 27 tok/s | tight |
| M4 Pro 24GB | 24GB | 27 tok/s | tight |
| A100 PCIe 80 GB | 80GB | 274 tok/s | comfortable |
| H100 SXM5 80 GB | 80GB | 551 tok/s | comfortable |
| GeForce RTX 3090 | 24GB | 132 tok/s | comfortable |
| Radeon RX 7900 XTX | 24GB | 109 tok/s | comfortable |
| GeForce RTX 4080 | 16GB | 106 tok/s | comfortable |