Runpod GPU Cloud published hourly pricing (Community and Secure Cloud)

Vendor pricing page, marked kind:blog per the repo rule that a vendor page restating its own list price is a vendor claim rather than an independent measurement. Used only as the counterfactual in the derived cost-per-token arithmetic - what a buyer pays per GPU-hour versus what the electricity alone costs to run that GPU-hour. Page states two tiers and labels them 'Per hour' and 'Per second'. Observed 2026-10-03: Community Cloud per GPU-hour H200 141GB $4.59, B200 180GB $6.79, B300 288GB $7.89, RTX Pro 6000 96GB $2.09, H100 NVL 94GB $3.19, H100 PCIe 80GB $2.89; Secure Cloud per worker-hour H200 $5.93, B200 $8.64, B300 $9.98, RTX 6000 Pro $3.49. Community figures come with attached CPU and RAM (the page lists e.g. H200 with 276 GB RAM and 24 vCPUs), which is why the per-worker and per-GPU-hour figures are not interchangeable. Pricing pages change without notice and several regions differ, so the derived cost record carries the observation date and treats this as an asking price rather than a realised one. 2026-10-11: Re-read RunPod pricing page. Confirms published asking hourly prices. Retained at confidence 0.5 as fluctuating vendor list price.