AI model pricing table
Compare model input and output pricing across providers.
Tokenscost
GPU Rental Price Index — Cheapest Cloud GPUs
Track hourly GPU rental prices across cloud providers and find the cheapest instance for training and inference.
AI API providers
Browse provider profiles and API pricing context.
Static pricing data
Fetch the public pricing JSON feed without executing JavaScript.
Last updated: · Source: aws pricing · azure pricing · gcp pricing · runpod pricing · vastai pricing · lambda pricing · coreweave pricing · nebius pricing · crusoe pricing · cudo pricing · oracle pricing · digitalocean pricing · voltagepark pricing · jarvislabs pricing · fly pricing · bitdeer pricing · denvr pricing · exabits pricing · taiga pricing · qarnot pricing · ori pricing · Methodology
Frequently asked questions
What do the GPU prices measure?
The recorded snapshot lists 5476 instances in USD per hour, grouped by provider, accelerator, GPU count, region and pricing tier. These are stored observations, not a guarantee that capacity is available; compare like-for-like instance configurations.
Are GPU prices quoted per token?
No. GPU rentals are billed per instance-hour. Normalize multi-GPU instances by GPU count for per-GPU-hour comparisons. Per-token cost also depends on throughput, utilization, model size, memory and operational overhead; do not compare an hourly rate directly with an API token rate.
How often is GPU pricing updated?
The GPU file records a weekly cadence and was last updated 2026-11-01. Rental-history observations are monthly. Opening this page does not verify or refresh provider quotes; the recorded snapshot date stays unchanged until data changes.
What additional GPU charges can apply?
Storage, data transfer, networking, taxes, minimum commitments and spot interruptions may affect the total bill. Reserved, spot and community tiers have different terms. Follow the linked provider sources and use your expected utilization before choosing a rental option.