GPU rental price data: downloads, API and citation
Every price on this site, free to download and reuse with credit: 469 offers from 37 providers across 40 GPU models today, and a daily, append-only record kept since 5 July 2026 (100 days). No account and no key.
Downloads
| What | Format | Contents |
|---|---|---|
| Today's prices | CSV | One row per offer, the eight fields below. |
| Today's snapshot | JSON | Every offer with all published configuration fields (node size, region, stock, CPU, RAM, commitment), plus each provider's check status. |
| Collection health | JSON | Snapshot date, offer and provider counts, and which providers' checks failed. |
| Archived history | Zenodo | The dataset's permanent record and versions, DOI 10.5281/zenodo.21435394. |
The JSON endpoints allow cross-origin requests, so a page or notebook can read them directly. Please cache: the data changes once a day, after the morning collection.
What each field means
date- The day of the price check (UTC) the row belongs to.
provider- The provider's identifier, as in the provider page address (runpod, lambda, voltagepark…).
gpu- The exact GPU variant, from our registry (h100-sxm, a100-pcie-80, rtx-4090…).
vram_gb- Memory per GPU in GB, as the provider lists it.
usd_per_hr- The price per GPU per hour in US dollars, before storage, data transfer and tax.
kind- Billing type: on-demand, secure (a provider's datacenter tier), community, spot, marketplace, reserved or serverless.
source_url- The provider's own page the price was read from.
fetched_at- When we read it (ISO 8601, UTC). A provider whose check failed keeps its last date; it is never shown as today's.
A price is per GPU. Where a provider sells only whole nodes, the snapshot's min_gpus gives the smallest node and the node price is the per-GPU rate times that count; where a provider publishes no node size, the field is absent rather than assumed to be one. Stock is recorded only where a provider reports it and is never inferred from a price. How we collect, check and rank is set out in the methodology.
Licence and citation
The data is free to use, share and adapt under Creative Commons Attribution 4.0: credit "GPU Rental Prices" with a link. To cite it:
GPU Rental Prices. (2026). GPU rental price index [Data set]. gpurentalprices.com. https://doi.org/10.5281/zenodo.21435394 (Updated 2026-10-12)
BibTeX
@misc{gpurentalprices_2026,
title = {GPU Rental Price Index},
author = {{GPU Rental Prices}},
year = {2026},
doi = {10.5281/zenodo.21435394},
howpublished = {\url{https://gpurentalprices.com/data}},
note = {Updated 2026-10-12}
}Badges
Each GPU has a live price badge for a README or a dashboard, for example https://gpurentalprices.com/badge/h100.svg, which shows the cheapest on-demand H100 rate and links back to its page when you wrap it in a link.
Questions about the data, a correction or a bulk need: write to [email protected].