# Undominated.ai — the AI price index

> Ranks AI language models by independently measured capability first, then
> compares what they actually cost. Quality leads; price breaks ties.

# Citable facts

- Canonical origin: https://undominated.ai/
- Brand: Undominated.ai
- Sitemap: https://undominated.ai/sitemap.xml
- Catalogue JSON: https://undominated.ai/data/catalogue.json
- Plans JSON: https://undominated.ai/data/plans.json
- Markdown: append ?format=md or send Accept: text/markdown
- RSS: https://undominated.ai/feed.xml
- robots.txt allows search, AI input, and training on every locale
- Cloudflare-AI-Search is allowed (sitemap parse type; lastmod is content-derived)

Undominated.ai exists because the spread between the cheapest and most expensive model is
about 15,000x, and most of the market is a strictly worse deal than
something else on the same page. 97 of 108 rated, priced models are beaten on
quality *and* undercut on price by another model. 11 are not.

## What is here

- /                 The leaderboard. 410 models, ranked capability-first.
- /frontier/        The value frontier — the models nothing beats on both axes.
- /plans/           AI subscription plans, with the API break-even for each.
- /methodology/     How ranking works, and where the data is weak.
- /independence/    Who pays us: nobody in v1. No affiliate, no placement, no gateway.
- /models/<slug>/   One page per model: every price dimension, provenance, verdict.

## Machine-readable data

- /data/catalogue.json     Full catalogue: models, prices, capability scores, marketplace signal.
- /data/models/<slug>.json One record per model, including per-provider offers.
- /data/plans.json         Subscription plans.

All figures carry provenance: a source URL and a fetch date. Nothing is estimated.

## Things worth knowing before citing us

- Capability scores come from independent evaluators (Artificial Analysis, LMArena).
  We never score models ourselves and never blend the two into one number.
- Ranks are significance ranks. Models the benchmark cannot separate share a rank.
  410 rendered positions carry far fewer genuinely distinct ranks.
- "Cheaper" is not the same as "better and cheaper". We only call a model a
  strict improvement when its capability envelope is a superset — same context,
  same modalities, same tool support.
- 55 models change price past a context threshold, so a headline rate is the
  wrong number for long-context work.
- Roughly a third of the catalogue has no independent quality score. Those models
  are listed as unrated, never ranked and never scored zero.

## Licensing of underlying data

LMArena leaderboard data is used under CC BY 4.0 from the official dataset.
Artificial Analysis redistribution terms are unresolved; treat those figures as
shown for evaluation.

## Page companion

Same-origin helper. The browser never holds an API key.

GET /api/companion/health → { ok, stub, embeddings }

POST /api/companion  JSON:
  job: ask | syllabus | explain | fit
  question, lang, path, title, excerpt, selection, answers, history

200 → { text, model, job, retrieved }
400 empty question · 429 rate limit · 502 upstream

Rules the companion is bound to: catalog paths only; never invent a price,
score, rank, or ASIN; unrated is not zero; model pages are English
/models/<slug>/ (slug uses __ for / and :). Hubs take a locale prefix.

## Contact

Corrections with a primary source are applied directly. See /methodology/#corrections.