Models · Qwen · live prices
Qwen 2.5 72B: GPUs needed and cost to self-host
72.7B parameters. In FP8 that is about 87 GB with room for the KV cache.
Memory, FP8
87 GB
weights + ~20%
Cheapest per 1M output tokens
$0.23
2× A100 · on-demand · 60% utilization
Hosted API, sample list price
$1.20
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 2× on Vast.ai · on-demand · $0.67/hr · ~1,346 tok/s $0.23per 1M out
- H200141 GB 1× on Vast.ai · on-demand · $1.32/hr · ~1,585 tok/s $0.39per 1M out
- 509032 GB 4× on Vast.ai · on-demand · $1.83/hr · ~1,420 tok/s $0.60per 1M out
- 409024 GB 8× on Vast.ai · on-demand · $2.49/hr · ~1,597 tok/s $0.72per 1M out
- H10080 GB 2× on RunPod · community · $3.98/hr · ~2,212 tok/s $0.83per 1M out
- B300288 GB 1× on RunPod · community · $6.94/hr · ~2,641 tok/s $1.22per 1M out
- RTX PRO 600096 GB 2× on Vast.ai · on-demand · $2.00/hr · ~710 tok/s $1.31per 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 | BF16174 GB | FP887 GB | INT444 GB |
|---|---|---|---|
| H10080 GB | 4× | 2× | 1× |
| H200141 GB | 2× | 1× | 1× |
| B300288 GB | 1× | 1× | 1× |
| A10080 GB | 4× | 2× | 1× |
| RTX PRO 600096 GB | 4× | 2× | 1× |
| 509032 GB | 8× | 4× | 2× |
| 409024 GB | — | 8× | 4× |