Methodology

How this catalogue is collated, how often, what the confidence levels mean, and how to tell us we are wrong.

Where the data comes from

Several sources, because none of them is complete on its own:

The confidence levels

Every published field carries one, and it is always visible.

unverified
Collected, but nobody has checked it against a primary source. Most of the long tail sits here, and it is published anyway — a comparison site that omitted everything unchecked would be less useful than the open seed it started from. The flag is the point.
ai-verified
Read from a first-party source by machine. Strong, but no human has confirmed it.
human-verified
A named person checked it against a primary source, on a recorded date. Nothing automated can produce this level — not a consensus of sources, not a high-confidence extraction.

Prices

Published in the provider's own currency. We never convert and we publish no exchange rate, because a converted price is two claims — the rate and the date — presented as one. Where a model is quoted in more than one currency, comparison tables say so rather than ranking across them.

Prices are list prices. A negotiated rate is a private matter between a customer and a provider and appears nowhere here.

Cadence

Ingestion and publication run daily. Every published document carries the date it was built, and a model's page shows when its offering last changed.

Telling us we are wrong

A wrong price or a missing capability is worth reporting — corrections go into the same review queue as every other proposed change, and a correction with a link to the provider's own page will usually be applied the same day. Open an issue on the platform repository.