How the board prices an AI media job
The standard job per category, how the median across providers is computed, which prices count, freshness, and a glossary of every term on the board.
The standard job
Each category has one standard job, so prices compare: video is a 720p, 5-second clip and is shown per second; images are one 1K image; speech is 1,000 characters; music is a 30-second track. A model with no such setting is priced at its closest one.
The median
A model's price is the median of the counted providers' prices for its standard job. A price counts when it is public, sourced, recent enough for its kind of source, and not for a deprecated offering.
Official prices and who sells them
When a maker publishes a self-serve API price, that price is curated from its own page: the URL, the day it was checked and the words it was read from. It votes once in the median. A contract-only price is shown and does not vote. Official prices joined the median on 16 October 2026; earlier history is not backfilled. Every listed price names its source page, a link and the day it was checked. Our tables, medians, rankings and wording are CC BY 4.0: use them for anything, commercial included, with the attribution "Model Price Track (modelpricetrack.com)" and a link. Third-party prices are listed as published by each provider, with a link, and are not relicensed. Arena quality scores stay CC BY 4.0 with Arena's attribution.
Official is the maker on its own API. Platform is an API host that sells many makers, or a maker API selling another maker's model. Reseller is an app or subscription that sells generations.
% vs official is how far a provider's price sits from the maker's, on the same variant and standard job. Within ±1% it reads "= official". Green is cheaper, red is dearer. It shows only while the official price is fresh and self-serve.
The Sold by filter recomputes the median over the channels you pick (official, platform, reseller). Filtered views keep a ?channel= in the URL.
Quality scores
Quality scores come only from the Arena leaderboard dataset on Hugging Face (CC BY 4.0). The site never fetches arena.ai. Each score names Arena, links to the dataset, and shows the leaderboard publish date. A score is fresh for 60 days from that date. Artificial Analysis is not shown. Value sort puts models on the price-quality frontier first, then the rest by Elo gap to that frontier. Arena does not endorse that ranking.
Every number is in the open read API, documented on the API page with its license and how to cite it.
Freshness
1 of 2 provider price APIs answered on the last read, 17 Oct 2026. Curated prices last checked 17 Oct 2026.