GPU cloud and LLM pricing data for journalists, researchers, and analysts.
Compute Comparison is an independent tracker of GPU cloud and LLM inference pricing. We poll provider APIs and public pricing pages every 15 minutes. Observed update cadence varies by provider; the live tables show a last-observed timestamp per row. Data is available for editorial and research use under CC BY 4.0 with attribution. API access requires a key and is subject to rate limits.
What the dataset includes
- Hourly on-demand and spot GPU prices by provider and chip
- LLM per-token pricing including cached-input rates
- Daily price-history snapshots, permanently retained
- Benchmark data: FP16/BF16 TFLOPS, memory bandwidth, and inference throughput
How prices are checked
Prices are collected via automated fetches from provider APIs and public pricing pages on a 15-minute polling cycle. Each fetch is normalized to a common schema and stored with a timestamp and source identifier. Observed update cadence varies by provider.
Rows are reviewed against official provider sources on a rolling basis. When a price is found to differ from the provider's published rate, the corrected value is labeled as a correction in the dataset. We do not remove historical rows — corrections are appended alongside the original.
How to request data
Email hello@computecomparison.com with the providers, chips or models, and time window you need. Custom data pulls for journalists are free for editorial use.
Historical price data is available under CC BY 4.0 with attribution to Compute Comparison and a link to computecomparison.com.
For full details on collection methodology, normalization rules, and known data limitations, see the methodology page.
For interview requests, data questions, or press inquiries, reach out directly. We respond to journalists and researchers promptly.
hello@computecomparison.com