Qwen1.5 models
18 Qwen1.5 models on Hugging Face, from 464M to 72.3B parameters. At the precision each one is published in, the smallest needs about 1.4 GB of VRAM (Qwen1.5-0.5B, cheapest live fit: RTX 5060 Ti) and the largest about 162 GB (Qwen1.5-72B, cheapest live fit: RTX A5000). The cheapest way to run Qwen1.5-0.5B is $0.110/hr.
Qwen1.5 models
| Model | Parameters | Published as | VRAM needed | Live GPU fit | Est. $/hr |
|---|---|---|---|---|---|
| NuExtract-tiny | 464M | F32 | 2.1 GB | V100 | $0.088/hr |
| Qwen1.5-0.5B | 620M | BF16 | 1.4 GB | RTX 5060 Ti | $0.110/hr |
| Qwen1.5-0.5B-Chat | 620M | BF16 | 1.4 GB | RTX 5060 Ti | $0.110/hr |
| Qwen1.5-1.8B | 1.8B | BF16 | 4.1 GB | RTX 5060 Ti | $0.110/hr |
| Qwen1.5-1.8B-Chat | 1.8B | BF16 | 4.1 GB | RTX 5060 Ti | $0.110/hr |
| Qwen1.5-4B | 4.0B | BF16 | 8.8 GB | RTX 5060 Ti | $0.110/hr |
| Qwen1.5-4B-Chat | 4.0B | BF16 | 8.8 GB | RTX 5060 Ti | $0.110/hr |
| CodeQwen1.5-7B-Chat | 7.3B | BF16 | 16.2 GB | RTX A5000 | $0.176/hr |
| Qwen1.5-7B | 7.7B | BF16 | 17.3 GB | RTX A5000 | $0.176/hr |
| Qwen1.5-7B-Chat | 7.7B | BF16 | 17.3 GB | RTX A5000 | $0.176/hr |
| Qwen1.5-14B | 14.2B | BF16 | 31.7 GB | RTX 4080 Super | $0.338/hr |
| Qwen1.5-14B-Chat | 14.2B | BF16 | 31.7 GB | RTX 4080 Super | $0.338/hr |
| Qwen1.5-MoE-A2.7B | 14.3B | BF16 | 32.0 GB | RTX 4080 Super | $0.338/hr |
| Qwen1.5-MoE-A2.7B-Chat | 14.3B | BF16 | 32.0 GB | RTX 4080 Super | $0.338/hr |
| Qwen1.5-32B | 32.5B | BF16 | 72.7 GB | A100 | $1.31/hr |
| Qwen1.5-32B-Chat | 32.5B | BF16 | 72.7 GB | A100 | $1.31/hr |
| Qwen1.5-72B | 72.3B | BF16 | 162 GB | RTX A5000 × 7 | $1.23/hr |
| Qwen1.5-72B-Chat | 72.3B | BF16 | 162 GB | RTX A5000 × 7 | $1.23/hr |
VRAM is for the precision the model is published in, with the same overhead and an 8,192-token context assumed on every page; see the methodology. The fit shown is the lowest-priced single GPU type that holds the model at that precision, or the lowest-priced multi-GPU set (up to 8) when none does.
