Pass IndexThe State of AISign in

Every list

231 lists, each one a way of cutting the catalogue that somebody would actually search for. Any two of the first three can be crossed — open-weight voice models, coding models that fit in 32 GB — and every crossing that has a member has its own address.

Chosen rather than filtered 6 lists

Lists nobody could derive from a column. Which model is best at each kind of work, which of the measured ones you can run for almost nothing, what is given away, what fits on what, and what runs where the host cannot read your prompt.

What a thing is for 20 lists

The job somebody wants done. Read off the modality, the maker's own description and the boards a thing is measured on — never guessed at.

What you may do with it 4 lists

Whether the weights are published, and what the licence asks in return. Every one read off a model card rather than inferred.

What your own machine can hold 9 lists

Open-weight models small enough to run yourself, by the memory of the device. Four-bit weights, total parameters counted, arithmetic stated on each page.

What goes in and what comes out 37 lists

The shape of a thing, which decides whether it can be used at all. Lists of fewer than five are left off this map but still answer.

Everything, in order 6 lists

The whole catalogue, a hundred to a page, alphabetically.

Around the market 2 lists

The people behind the companies, and the wire the crawler reads.

Who is measured, and where 147 lists

Public leaderboards the catalogue reads. The count is how many of the models each board ranks are things the catalogue also holds.