Methodology.

Every rate starts from the provider's own pricing page — never a blog post, never a guess. The full list lives on Sources.

Rates are re-fetched every few minutes into our database. The calculator reads live rows, so a provider price change flows through on its own — no app update needed.

Providers bill per million tokens, so we keep that unit: input, output, 5-minute cache write, 1-hour cache write, and cached input each carry their own price. Your monthly estimate is token allocation × rate × request volume.

Repeated context costs less. Pick a cache tier and cached versus fresh tokens are priced separately instead of averaged into one misleading number.

Your prompt is tokenized in your browser — nothing is uploaded. Real tokenizers differ slightly per model, so token counts are approximate while the per-million prices are taken verbatim from the source.

A rate counts as verified when it links back to a live provider page and was re-checked inside the polling window. Stale or sourceless rows never reach the calculator.