About ModelsAtlas

Every page answers its question in the first two lines, shows the number that answers it, and lets you open the source of that number. A price shows where and when we read it. A verdict names the commit we read. Where we do not know something we say so: “Not disclosed”, “None linked yet”, “Not advertised”.

We use “we” for ModelsAtlas and “you” for the reader. We do not name the data vendors we gather from: a figure from our own gathering reads “Our catalogue”, and vendor documentation is named and linked because it is the vendor speaking.

A research pipeline reads vendor documentation, pricing pages, registries and repositories, and records every fact with its source and the date it was retrieved. Proposed changes are checked and reviewed before they are published.

This site reads one set of records when it is built and renders them. It does not fetch facts at runtime, and a page template never corrects a record: if a price or a spec looks wrong, that is a data problem, and a correction is how it gets fixed.

Skills and MCP servers are read in full at a named commit and given a verdict, a grade and a score. The methodology says how, and how to check a result yourself.

No vendor pays for a position in a ranking or a verdict, and a BLOCK page stays public. How we make money lists what does pay for the site and where ads never appear.

Some things you may expect on a map like this are not here because we do not have the data behind them: price history, speed measurements, usage rankings, security advisories and deprecation dates. We would rather leave a section out than fill it with a guess.