Skip to content

How we rank

The criteria and order of the ranking. It cannot be bought.

Updated Nov 13, 2026

The order

We simulate your typical week against every plan and apply these criteria, in this order:

  1. First, it must cover your models. A plan that does not include every model you use cannot be the recommended one, even if it is cheaper. We allow a margin of 5% of your tokens. Plans that do not cover them rank behind those that do, from most to least coverage, and a plan that covers none goes last. We say so on every plan: “only covers X% of your tokens”.
  2. Then, no known blocks, cheapest to most expensive. Among the plans that cover your models and do not block you in your typical week, the one with the lowest monthly cost wins. If two cost the same (half a cent apart or less), the one that leaves more margin in the busiest stretch wins. A limit the provider does not publish neither raises nor lowers a plan: we mark it “Not published” and lower its confidence. If none of its limits has a figure, we cannot simulate anything and the plan goes behind those that do.
  3. Then, the ones that block. A plan blocks if your typical week touches one of its limits or if a monthly pool runs out with no overage payment. In that case, every day of use left in the month counts as a block. They are ordered from fewest to most blocks per week and, with the same blocks, by cost and margin. The block is shown, not hidden.
  4. And if everything ties, the identifier. The last criterion is the alphabetical order of the plan's identifier. It only exists so the order is always the same.

How we calculate cost

  • Subscription. Monthly price (the annual one, divided by 12) plus the overage you would pay if your usage goes past the included pool.
  • Pay per use. Your tokens at API price, only for the models that plan serves.
  • In euros, VAT included. We add Spanish VAT (21%) to everything published without VAT: prices in dollars, API rates and overage. Dollars are converted to euros at the ECB reference rate of Nov 5, 2026.
  • In dollars, without VAT. We remove VAT from the prices that include it (Google publishes its plans in euros with VAT).
  • What you would pay through the API. We compare your usage at API price with the plan's cost using the same VAT rule on both sides, so the ratio is the same in euros as in dollars. If a model has no published API price, we leave it out and say so.

The best fit

It is the first plan in the ranking if it covers all your models and does not block you. If there is none like that, we do not choose for you: we tell you it is time to combine plans.

Measured and estimated

Every figure in the ranking says whether it is measured (published or observed in your profile) or estimated. Estimates are shown as such and never presented as official data.

Confidence

Each plan carries a confidence (high, medium or low): that of the weakest figure it depends on. It is lowered by an estimated figure, a community figure, a limit that is not published or a model with no API price. Confidence does not move the order: the best fit is always shown with its own confidence and with what lowered it.

It cannot be bought

No provider pays to appear or to move up. A plan's position comes from the rule above and nothing else.

Affiliate links

There are none today. If there were, they would be declared next to the link, on this page and on independence, and would not change the order: the ranking is calculated first and without knowing whether a link is an affiliate link.