ModelCap policies

Editorial policy

Effective September 22, 2026

ModelCap sets out to answer a narrow question honestly: which AI models the public evidence shows are most capable, what they cost, and where you can use them. This policy explains how the rankings and the notes are made, who answers for them, and how mistakes are fixed.

Who is responsible

Model Cap LLC publishes modelcap.ai and is responsible for everything on it: the ranking method, the software that applies it, the notes, and every correction. Questions and complaints go to the address on the contact page, which the people who run the site read.

How the rankings are made

• Every figure comes from a named public source: independent leaderboards, publisher model cards and provider catalogues. The methodology lists each source and the role it plays.
• Versioned software computes the scores on a fixed schedule. Nobody places a model by hand, and no language model decides a number.
• A publisher's own reported results can place a new model, but they are labelled as publisher-reported and never count as independent measurement.
• Popularity, downloads, prices and provider counts are shown as context. They never raise a model's capability score.
• A model that cannot be tied to one exact catalogue identity gets no rank. A gap is better than a guess.
• The method is versioned. The methodology is the current contract, and the notes explain why a rule changed.

Independence

• No advertiser, publisher, provider or reader can buy, request or negotiate a rank, a badge or a mention. Messages do not move models; only admitted public evidence does.
• ModelCap does not run affiliate links on model or provider pages, and does not accept payment to add, remove, rename or re-rank a model.
• Advertising, when enabled, is served by Google in fixed, labelled positions outside the rankings, and it has no say in what the board shows.
• The release radar and prediction-markets pages report third-party market prices as context. They are kept apart from the board and never feed a score.

Notes and the use of AI tools

Notes are ModelCap's editorial record: what the board did on a given day, why a rule exists, and how to read a row. Each note is dated, lists the sources it relies on, and describes the board as it stood on that date. A note is revised only when a later change would make it misleading, and the revision is dated on the note.

ModelCap uses software, including AI assistants, to build the site and to help draft notes. No AI tool produces a ranking figure: the figures in a note come from the published data. Each note is reviewed before it is listed, and Model Cap LLC answers for it like anything else on the site.

Corrections

When a figure is wrong, we fix the rule or the reading of the source that produced it rather than editing one number, so the same mistake cannot recur for the next model. Report an error through the contact page with the model, the source you are comparing against, and what you expected to see.

Changes to the method are recorded in the notes and the methodology. A note found to be wrong is corrected in place and carries the date of the correction.