Models · Qwen · live prices
Qwen 2.5 32B: GPUs needed and cost to self-host
32.5B parameters. In FP8 that is about 39 GB with room for the KV cache.
Memory, FP8
39 GB
weights + ~20%
Cheapest per 1M output tokens
$0.10
1× A100 · on-demand · 60% utilization
Hosted API, sample list price
$0.80
per 1M output tokens
Cost per 1M output tokens, FP8
Prices observed 12m ago. Cheapest on-demand offer per GPU, whole instances, 60% average utilization.
- A10080 GB 1× on Vast.ai · on-demand · $0.34/hr · ~1,506 tok/s $0.10per 1M out
- H200141 GB 1× on Vast.ai · on-demand · $1.32/hr · ~3,545 tok/s $0.17per 1M out
- 509032 GB 2× on Vast.ai · on-demand · $0.92/hr · ~1,588 tok/s $0.27per 1M out
- 409024 GB 2× on Vast.ai · on-demand · $0.62/hr · ~893 tok/s $0.32per 1M out
- RTX PRO 600096 GB 1× on Vast.ai · on-demand · $1.00/hr · ~1,323 tok/s $0.35per 1M out
- H10080 GB 1× on RunPod · community · $1.99/hr · ~2,474 tok/s $0.37per 1M out
- B300288 GB 1× on RunPod · community · $6.94/hr · ~5,908 tok/s $0.54per 1M out
The tick marks the sample hosted-API price. Estimates: tokens/s from memory bandwidth with batched serving, not a benchmark. Idle hours still bill, so lower utilization means pricier tokens.
GPUs per replica by precision
One node, split 1, 2, 4 or 8 ways; 90% of VRAM usable.
| GPU | BF1678 GB | FP839 GB | INT420 GB |
|---|---|---|---|
| H10080 GB | 2× | 1× | 1× |
| H200141 GB | 1× | 1× | 1× |
| B300288 GB | 1× | 1× | 1× |
| A10080 GB | 2× | 1× | 1× |
| RTX PRO 600096 GB | 1× | 1× | 1× |
| 509032 GB | 4× | 2× | 1× |
| 409024 GB | 4× | 2× | 1× |