Press & data kit

Price of Compute is the free reference for what compute costs: GPU rental prices across 16 cloud providers normalized to USD per GPU-hour (collected hourly, published as cross-provider daily medians), verified LLM token prices for 20 models, and a composite index. Methodology public, history free to January 2023, no login anywhere. Built and run by Noah Flanary.

Numbers you can print today

• The Price of Compute Index: 116.9 (daily close, base 100 = 2026-08-10) · live print 122.4
• NVIDIA H100 SXM median: $3.87/GPU-hr across 11 providers ($1.99–$11.68)
• Widest same-GPU spread today: NVIDIA H100 SXM at 5.9× between cheapest and priciest provider
• History depth: daily series to 2023-01 on covered SKUs (archival reconstruction from dated Internet Archive snapshots, labeled, never estimated)

Scheduled price events we will record

• 2026-09-01: Claude Sonnet 5 intro pricing ends ($2/$10 → $3/$15 per MTok) — a rare published price increase, timestamped on our events feed.
• Date TBA: DeepSeek has pre-announced a “significant” API price increase in its own docs — we track the before/after.

Embeds & assets

Live badges (hotlink freely): priceofcompute.com/api/badge/index.svg and /api/badge/h100-sxm.svg (any SKU slug). OG chart cards: /og?sku=h100-sxm. Logo: /logo.svg · lockup. Full data via the free API (JSON + CSV; 1,000 req/day/IP).

How to cite

Prose: “according to Price of Compute (priceofcompute.com)…”. Everything is free with the attribution line “Data: Price of Compute — priceofcompute.com”. Formats incl. BibTeX: /cite. Rules, changelog, and every excluded source with its reason: /methodology.

Contact

Contact — data pulls, custom cuts, and interviews are fast; every observation stores its raw source payload, so any number can be traced end-to-end on request.