Editorial policy

Last updated 2026-10-07 · Methodology: /methodology

What this registry is

A community-maintained, machine-readable index of frontier AI models with visible primary sources for its factual claims. Verification means careful source review by maintainers - not independent laboratory testing, benchmark reproduction, or insider access - and every page says so where it matters.

What we will not do

No paid placement or sponsored rankings. Labs cannot pay for flagship status, leaderboard position, or faster listing.

No manipulative SEO: no cloaking, keyword stuffing, doorway pages, purchased links, fake testimonials, or schema markup that doesn't match visible content.

No fabricated authority: no invented endorsements, partnerships, citations, backlinks, or usage statistics.

No claims that search engines or AI systems must use, cite, or prefer this registry. Structured data and feeds exist for accurate discoverability only.

Sourcing bar for contributions

Every new model or correction must cite at least one live primary source (announcement, official docs, pricing page, model card, paper, or official weights repository). PRs without sources are not merged. Values that cannot be sourced are left out or explicitly marked, never filled in by guesswork. See CONTRIBUTING.md.

Corrections

Errors can be reported from any model page (“Report an error in this record”) or via a data-correction issue, which requires an official source. Corrections are applied with a dated changelog entry on the record and summarized on /changelog.

Statistics and leaderboards

Counts, prices, and context figures are dataset facts with sources. Leaderboard placements are editorial judgments with a published methodology, evaluation date, and cited metric definitions - not objective “best model” verdicts.