LLM cost calculator

What it does

Describe one request — how many input tokens it sends, how many of them the provider reads from its cache, how many output tokens come back — and how many requests you make a day. The calculator prices that workload on every model in Salyro's catalogue and lists them cheapest first: per request, per day and over the number of days you choose.

Everything runs in your browser. Nothing you enter is sent anywhere, and a shared link holds only numbers.

How it computes

• Prices come from Salyro's own price list: the rates, in dollars per million tokens, that Salyro records usage against. The page is rebuilt from that list, so it cannot disagree with it.
• Arithmetic is the function Salyro's usage records are priced with, calculateCost, run here unchanged. It works in exact decimals and rounds once, at ten places — never in floating point.
• A day and a period are the per-request amount times the number of requests. That is exactly how a ledger that prices every request on its own adds up.
• Price bands. Some models charge more per token once a prompt is long enough. The band is chosen from the request's input tokens, the way the engine chooses it, and the table says which band applied and why — for example Input over 272,000 tokens → higher band.
• Context windows and output limits come from the provider's own model pages. A request is flagged, and left out of the cheapest pick, when its input does not fit, when input plus output does not fit, or when its output is more than one answer may be.
• Announced prices. When a provider has announced a price change that the catalogue already holds, you can price the workload on the day it takes effect.

What it does not know

• A model with no price for part of the request is not priced, never $0. The most common case is cached input on a model whose price list states no cached rate: the calculator does not substitute the ordinary input rate.
• Context windows that could not be read are unknown. Where a provider's page could not be read when the facts were compiled, the calculator says so rather than taking a figure from somewhere else.
• It prices tokens, not requests: it does not know about per-request fees, image inputs, tool calls a provider bills separately, or taxes.
• It does not know your discounts, credits or committed-use agreements.
• Prices are as of the revision named under the table, not live.

FAQ

Why is a model listed as "not priced"?

The table gives the reason on each row. Usually the workload uses something the model's price list does not state a rate for — cached input, for example — or no price is in force on the date you chose. A missing price is shown as missing, because $0 would be a claim that the request is free.

Why does the price of a long prompt jump?

Some providers bill every token of a request at a higher rate once its input passes a threshold, not only the tokens past it. The band column shows the threshold the provider states.

Are these the prices Salyro charges?

They are the providers' list prices as Salyro records them, and the arithmetic is the one Salyro prices usage with. Your own bill depends on your agreement with each provider.

Can I share a result?

Yes. The link holds the numbers you entered, compressed after the #, which your browser never sends to a server.

Sources

The provider pages each price was read from, and when, are listed under the table: the catalogue and price-list revisions, and a link to each provider's pricing page. Context windows name the model page they were read from.