How the numbers are made
A job, not a list price
Every comparison starts from a concrete job: “one 8-second 720p clip with audio”, “one 1K image”, “a chat workload at 3 input tokens per output token”. For each route we first check it can do the job, then apply the matching price rule, convert credits to dollars, and add fees you must pay.
Exact, approximate, not comparable
Exact: the route does exactly what you asked. Approximate: it can do the job with a difference we spell out, such as “only makes 8-second clips”; it is priced at what it actually produces and listed separately. Not comparable: the platform only publishes a “from” price, or no price for that combination. Those rows are shown but never ranked.
Estimates
Some image models are settled by tokens. Where the platform gives a per-image figure we show it with ≈ and say so.
Discounts and fees
List price is what a new, ordinary account pays, including promo prices shown publicly. “Best public tier” uses the best published membership price or top-up bonus, and says when upgrade rules are not public. OpenRouter’s 5.5% top-up fee is included in landed cost.
Reference price
Official when the model’s maker publishes an API price. Market when several official partner platforms list the same price. We do not use platforms’ own “official price” labels; some were not official.
Routes
Official: the maker’s own endpoint. Official partner: a hosted endpoint run with the maker. Third-party: a route the platform itself labels as non-official. Not disclosed: the platform does not say.
What we do not measure yet
Success rate, speed and output quality. Testing with paid generations is the next step. Until then, the ranking is price only, and a cheap route may be unreliable.
Independence
No partnerships with any platform, no paid placement, no referral links. Prices were checked by hand on 2026-11-07 against each platform’s public pages.