Methodology

Where the data comes from

Inference Hub aggregates pricing and model availability from three kinds of source, in order of preference:

• Machine-readable provider feeds — where a provider publishes a live models/pricing endpoint (for example, OpenRouter's public API), we read prices directly from it.
• Provider pricing pages — for providers without a public feed, we extract pricing from their official pricing and documentation pages.
• Official announcements — launch posts and docs for context such as context windows, licensing, and capabilities.

How updates work

We run a reconciliation process that compares freshly collected prices against what we already have, records the differences, and updates the database. Every change to a provider's price or a model's specs is written to a public, dated log — you can browse it on the changelog. Each model and provider page also shows when its pricing was last updated.

Pricing conventions

• Token pricing is shown in USD per 1 million tokens, split into input and output where providers price them separately.
• Image models use price per image; video and audio use price per second; some endpoints use per-request pricing.
• Where a provider bills only by subscription or in a non-USD unit (for example, a token-denominated or credit-based model), we note that in the price field rather than inventing a per-token figure.
• Prices reflect standard, on-demand, public rates — not negotiated enterprise, committed-use, or batch discounts unless explicitly labeled.

Accuracy & corrections

AI pricing moves fast and providers change rates without notice. We update continuously, but a figure can briefly lag the source — always confirm against the provider before committing to production spend. Spotted something wrong or out of date? Email info@miranext.net and we'll correct it.

Use the data

All of this is queryable through our public API. If you use Inference Hub data in your own work, a link back to the relevant model or provider page is appreciated.