Methodology

Data collection

Pricing is scraped weekly (Monday 03:00 UTC) using Playwright (headless Chromium). Each scraper navigates to the provider's pricing page, waits for content to render, extracts GPU model names and per-hour prices, maps names to canonical slugs, and validates output against a Zod schema.

Rate limiting

Scrapers throttle to 1 request per 2 seconds. User-Agent: pricegpu.com/1.0 (+https://pricegpu.com/about).

Data accuracy

Prices reflect publicly listed on-demand rates at scrape time. Actual billing may differ due to promotions, spot pricing, minimum billing periods, or prices changing between weekly scrapes. Every table shows "as of [date], verify on provider site".

Open data

Raw price dataset at /api/prices.json under CC-BY-4.0. Scraper source code in the public GitHub repository.

Providers tracked

RunPod · Vast.ai · Lambda Labs · Paperspace · Modal · CoreWeave · TensorDock · Genesis Cloud · Hyperstack · Together AI · Replicate · Fal.ai · Salad · FluidStack · DataCrunch