How to read a model price list

Provider price lists quote two figures per model: a rate per million input tokens and a rate per million output tokens. Output is almost always the expensive side, typically three to eight times the input rate, because generating text is more compute-intensive than reading it.

That asymmetry is why the same model can be cheap for one team and expensive for another. A retrieval-heavy assistant that reads long documents and replies briefly is dominated by input cost. A code generator that reads a short prompt and writes a long file is dominated by output cost.

Blended price

To compare models on one axis we use a blended price: 75% input and 25% output, weighted by the list rates. It is a convention, not a law, and it favours models with cheap input. If your mix is different, the cost calculator on the pricing page lets you plug in your own volumes.

What the list price leaves out

Cached-input discounts (often 50–90% off repeated prefixes), batch-processing tiers and negotiated enterprise rates can all sit well below the list price. None of them are reflected in the figures on this site, so treat our numbers as a ceiling rather than a quote.

Free-tier models are listed at zero, which distorts any ranking that sorts purely by price. The value rankings on the pricing page therefore separate free routes from paid ones.