Undominated.ai

Press kit

Everything here is free to use with attribution. The figures below are read from the same catalogue the site publishes from, so this page cannot quote a number the site does not.

In one line

Undominated.ai ranks language models by independently measured capability first, then compares what they actually cost, across every provider that sells them.

In a paragraph

Undominated.ai is a price-comparison index for AI inference. It takes capability scores published by a public benchmark under an open licence, attaches them to the models you can actually buy, and shows what each one costs across providers at a workload you choose. It refuses to rank a model nobody has measured: an absent score is shown as absent, never as a zero. It takes no cut of inference, carries no affiliate links, and sells no placement.

The numbers, as of 19 Eyl 2026

429
models in the catalogue
52
providers
137
carry an independent score
6
on the value frontier

These are read from the live catalogue at build time. If they disagree with the site, the site is right and this page is stale — tell us.

Logo and marks

SVG, and the only files we ask you to use for the mark itself. The mark may be placed on any background; the lockups are drawn for dark grounds. Please do not recolour, rotate, or add effects to the mark, and please leave clear space around it equal to the height of the counter.

Campaign stills

Material Witness is the primary campaign family. Proof Cinema and Evidence Editorial are supporting. These are scene treatments around the Dual Witness mark — not a second logo. Use with attribution. Do not redraw the mark from the photograph.

Material Witness

Proof Cinema

Evidence Editorial

What you may quote, and what we would rather you did not

Safe to quote

  • Any figure on this page or on the site, with the date. Every published number is traceable to a source row, and the site says which.
  • That 137 of 429 models in this catalogue carry a score from the benchmark we use, and that the rest are shown as unrated rather than ranked low.
  • That the site takes no cut of inference and carries no affiliate links. See Independence.

Please do not write

  • “The leaderboard is wrong.” We do not claim that. A rank is a position; the benchmark also publishes an interval around it, and we draw the interval at the same size as the rank. See Ranges.
  • “These models are tied” or “statistically indistinguishable”. Overlapping intervals do not establish equivalence, and the site is careful never to say they do.
  • “Unmeasured” or “unevaluated” for a model with no score here. It means no entry on this board on this date. Vendors publish their own results.

Sources and licensing

Capability scores are LMArena Elo, used under CC BY 4.0 from the official dataset distribution. Attribution travels with every derived figure. We do not scrape their site.

Prices come from providers’ own public APIs. They are facts a vendor published, reproduced with the date we read them.

One commercial benchmark we consult is not published anywhere on this site, because we have no redistribution licence for it. It informs what we check and never what you read.

Our own errors

We keep a public log of every wrong number we have published and how it was caught, including the ones nobody else noticed. It is the most useful page here for anyone assessing whether to trust the rest, and it is deliberately not hidden. See Corrections.

Contact

The fastest route is the public tracker — issues are answered in the open, which is the point. For anything that needs a person, open an issue and say so.

Public tracker