Models / Qwen3.5
Qwen3.5 by Qwen
9 models (0.8B to 397B) · ↓ 33.8M downloads a month · from $0.08/h on GTX 1660 SUPER 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.
Qwen3.5 2B
| Version | Runtime | Weights | GPU memory | Cheapest machine now |
|---|
| OriginalQwen/Qwen3.5-2B-Base | vLLM | 4.5 GB | 7.7 GB | $0.10/h RTX 3060 |
| InstructQwen/Qwen3.5-2B | vLLM | 4.5 GB | 7.7 GB | $0.10/h RTX 3060 |
Qwen3.5 27B
| Version | Runtime | Weights | GPU memory | Cheapest machine now |
|---|
| OriginalQwen/Qwen3.5-27B | vLLM | 56 GB | 66.3 GB | $0.89/h 2x Nvidia A40 |
| FP8Qwen/Qwen3.5-27B-FP8 | vLLM | 31 GB | 38.9 GB | $0.45/h Nvidia A40 |
| AWQ 4bit (cyankiwi)cyankiwi/Qwen3.5-27B-AWQ-4bit | vLLM | 20 GB | 26.9 GB | $0.34/h 2x RTX 5060 Ti |
| AWQ (QuantTrio)QuantTrio/Qwen3.5-27B-AWQ | vLLM | 22 GB | 28.9 GB | $0.34/h 2x RTX 5060 Ti |
Qwen3.5 35B
| Version | Runtime | Weights | GPU memory | Cheapest machine now |
|---|
| OriginalQwen/Qwen3.5-35B-A3B-Base | vLLM | 72 GB | 82.9 GB | $0.89/h 2x Nvidia A40 |
| InstructQwen/Qwen3.5-35B-A3B | vLLM | 72 GB | 82.9 GB | $0.89/h 2x Nvidia A40 |
| FP8Qwen/Qwen3.5-35B-A3B-FP8 | vLLM | 37 GB | 44.6 GB | $0.45/h Nvidia A40 |
| GPTQ Int4Qwen/Qwen3.5-35B-A3B-GPTQ-Int4 | vLLM | 24 GB | 30.1 GB | $0.41/h RTX PRO 4500 |
| NVFP4 (txn545)txn545/Qwen3.5-35B-A3B-NVFP4 | vLLM | 25 GB | 31.2 GB | $0.41/h RTX PRO 4500 |
Qwen3.5 122B
| Version | Runtime | Weights | GPU memory | Cheapest machine now |
|---|
| OriginalQwen/Qwen3.5-122B-A10B | vLLM | 250 GB | 281.1 GB | $5.33/h 8x Nvidia A100 |
| FP8Qwen/Qwen3.5-122B-A10B-FP8 | vLLM | 127 GB | 144.4 GB | $1.38/h 8x RTX 3090 |
| NVFP4 (nvidia)nvidia/Qwen3.5-122B-A10B-NVFP4 | vLLM | 84 GB | 95.9 GB | $1.38/h 8x RTX 3090 |
| AWQ 4bit (cyankiwi)cyankiwi/Qwen3.5-122B-A10B-AWQ-4bit | vLLM | 80 GB | 92 GB | $1.10/h 2x RTX 4090 |
Qwen3.5 122B
| Version | Runtime | Weights | GPU memory | Cheapest machine now |
|---|
| OriginalQwen/Qwen3.5-122B-A10B-GPTQ-Int4 | vLLM | 79 GB | 90.7 GB | $1.10/h 2x RTX 4090 |
Qwen3.5 397B
| Version | Runtime | Weights | GPU memory | Cheapest machine now |
|---|
| OriginalQwen/Qwen3.5-397B-A17B | vLLM | 807 GB | 899.8 GB | $44.03/h 8x Nvidia H200 |
| FP8Qwen/Qwen3.5-397B-A17B-FP8 | vLLM | 406 GB | 454.6 GB | $11.77/h 8x RTX PRO 6000 |
| NVFP4 (nvidia)nvidia/Qwen3.5-397B-A17B-NVFP4 | vLLM | 251 GB | 282.4 GB | $5.33/h 8x Nvidia A100 |
| NVFP4 V2 (nvidia)nvidia/Qwen3.5-397B-A17B-NVFP4-V2 | vLLM | 244 GB | 274 GB | $5.33/h 8x Nvidia A100 |
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").