What would your workload cost?
Set your volumes and every tracked endpoint is re-ranked by what it would actually bill you, cached input included. One row per model, priced at its cheapest host, with the range to its priciest: 98 of these 201 models are sold by more than one host, and the same model rarely has one price.
| # | Model · cheapest host | $/month | vs #1 | Same model, host range |
|---|
Building a whole app, not just the LLM line? The stack cost calculator adds hosting, database, email and payment fees on top — same prices, one bill. Method and sources: methodology; how token counts drift by tokenizer: the tokenizer tax.
How it's computed
Why is the bill a range and not one number?
The unit of record here is the endpoint, not the model: the same weights are served by several hosts at different prices. 98 of the 201 models in this calculator are sold by more than one host, so the same workload on the same model can produce very different bills. Each row ranks the cheapest host and shows how far the priciest one lands for your volumes.
How is the monthly bill computed?
Monthly bill = input × (1 − cache%) × input price + input × cache% × cached-input price + output × output price, in USD per million tokens. Where a host publishes no cached-input price, the cache share is billed at the full input price — no assumed discount, and the row says so.
Why do results differ from other calculators?
Most calculators list one price per model — the maker's — while this one prices every host of every model (750 endpoints for 201 models), so it often finds a cheaper host for the same weights. Free tiers billed at $0 are excluded, because a rate-limited quota is not a price. And a million tokens is not a fixed amount of text: every tokenizer cuts differently, so treat results as estimates — the tokenizer tax page quantifies by how much.
Are the prices current?
They are the index's cross-checked figures as of 2026-10-23T09:51:45Z, rebuilt with the whole site: the calculator embeds exactly the same data as the model pages, where every price carries its sources and verification date.