Independent testing

Accountability

Three boxes per company. What they say, what we checked, and what we reckon. Kept apart and labelled, so you are never a paragraph deep before working out which one you are reading.

Three kinds of statement. They are not the same thing.

What they say is their words, their link, and the date we read it. We are not nodding along. We are writing it down, so you can hold it against whatever they say next year.

What we saw only turns up once a person has gone and looked. Everything we mean to check and have not yet is listed underneath. You can see the homework we still owe you.

What we think is ours, and it has to show its working. If a view rests on nothing but a company's own account of itself, we say so. That is much weaker than it sounds.

None of this tells you whether a model is any good. We have not tested one yet. What you get for now is how each company writes things down — a smaller question, and one we can actually answer.

Anthropic

Registry last read it within the hour

What they say

Their words, their link, the day we read it. Quoting is not agreeing.

  • Publishes a machine-readable list of available models through its API, which is what this site reads rather than a marketing page.

    Their page · read 2026-09-29

  • Publishes a deprecations page listing models that are being retired and the dates.

    Their page · read 2026-09-29

What we have seen

Nothing yet. Nobody has been and checked any of it. So none of it counts as evidence, and we are not about to dress it up as any.

On the list. Not done.

  • Whether the model list can actually be fetched and parsed by a third party without special access.

    Our registry calls the address they publish, on a schedule, and writes down every success and every failure with the time.

  • Whether a model has ever disappeared from the list before appearing on the deprecations page.

    Compare the date a model leaves the fetched list against the date it is announced. The registry keeps both and never overwrites.

What we think

  • A machine-readable list and a dated deprecations page are the minimum a provider should manage, and not every provider does. That is a statement about their paperwork rather than their models. It is also the only sort we are in a position to make.

    This rests on what the company says about itself, and on nothing we have checked ourselves. That is a weaker thing than it sounds, and it is why there are not more opinions on this page.

OpenAI

Registry last read it within the hour

What they say

Their words, their link, the day we read it. Quoting is not agreeing.

  • Publishes a machine-readable list of available models through its API.

    Their page · read 2026-09-29

  • Maintains a deprecations page recording models that have been retired and their replacements.

    Their page · read 2026-09-29

What we have seen

Nothing yet. Nobody has been and checked any of it. So none of it counts as evidence, and we are not about to dress it up as any.

On the list. Not done.

  • Whether the model list can be fetched and parsed by a third party, and how often it changes.

    Our registry calls the address they publish, on a schedule, and writes down every attempt.

  • Whether a model name has ever quietly changed meaning, so the same name gives you something different.

    Our registry keeps every version of every name it has seen, with the date, and never paints over the old one.

What we think

Nothing. We have not watched OpenAI closely enough to give you a view worth your time. A view assembled from somebody’s press release is really just their press release.

Google

Registry last read it within the hour

What they say

Their words, their link, the day we read it. Quoting is not agreeing.

  • Publishes a machine-readable list of available models through its API.

    Their page · read 2026-09-29

What we have seen

Nothing yet. Nobody has been and checked any of it. So none of it counts as evidence, and we are not about to dress it up as any.

On the list. Not done.

  • Whether the model list can be fetched and parsed by a third party without special access.

    Our registry calls the address they publish, on a schedule, and writes down every attempt.

What we think

Nothing. We have not watched Google closely enough to give you a view worth your time. A view assembled from somebody’s press release is really just their press release.