Skip to main content
The core metric in the Ornn Data API is the price index: a number that tracks the on-demand rental price of a GPU model over time.

​
What the index value means

index_value is the going rate of compute for a GPU model, measured in US dollars per GPU-hour. It is a weighted average of all verified compute transactions across the market for that model. For example, an H100 SXM index of 2.00 means the market rate for an H100 SXM is about $2.00 per GPU-hour.
The index reflects actual transactions, not list prices. Treat index_value as a normalized market rate per GPU model rather than a quote from any single provider.

​
Hourly and daily

The API exposes the GPU index at two granularities: History endpoints automatically choose between hourly and daily: short ranges return hourly points, long ranges (more than ~30 days) return daily points. You can also force a granularity with the granularity parameter where supported.

​
Update frequency

  • GPU index (hourly / daily): refreshed hourly, on the hour.
  • Memory prices: refreshed once per trading day (weekdays). See Memory Price Index.
The last_updated (or updatedAt) field on each response tells you exactly when the value was computed.

​
Regions

The index currently tracks a single global region. The optional region parameter exists on several endpoints, but the only populated region is global, and it’s the default. You can omit it.
In some responses region echoes back as an empty string ("") and in others as "global". Both refer to the same global index. This is a cosmetic difference between endpoints, not different data.

​
Supported GPUs

The index covers these models: H100 SXM · H200 · A100 SXM4 · B200 · RTX 5090 · RTX PRO 6000 WS GPU names are case- and spelling-sensitive and must match exactly. Fetch the live list any time from /api/gpu-types, or see Supported GPUs & memory types.

​
Working with the endpoints

The GPU index is exposed through a family of endpoints. A few common starting points: A ranged daily history for the H100, oldest first:
GPU names go in the path and must be URL-encoded (H100 SXM becomes H100%20SXM). History ranges longer than ~30 days come back at daily resolution; shorter ranges come back hourly. Force a granularity with granularity where the endpoint supports it. See Analytics for what volatility and utilization mean.