Overview GPU Radar Free Where to run Free Token cost by GPU Free Capacity planner Free Tokens vs GPUs Free Idle cost Free Inference FinOps Free · Team Load testing Early access Synthetic monitoring Waitlist Settings Endpoints, data, API keys

Sign in free Back to the site

Token cost by GPU

What 1M output tokens of your model cost on each GPU, from live prices and estimated throughput, next to a hosted API.

Live prices

Model

Parameters (billions)

Precision

BF16 FP8 INT4

Average utilization 50%

Share of each GPU-hour spent serving. Idle hours still bill, so lower utilization means pricier tokens.

Include spot offers Cheaper, but can be interrupted

Loading GPU prices…

Couldn’t load GPU prices.

Estimate: output tokens/s from memory bandwidth with batched serving, cheapest offer per GPU billed in whole instances. Prefill (input tokens) is fast and not counted. Benchmark before you commit.