LLM cost calculator

Enter what you send and generate. The table below prices it on every current model, cheapest first — so you can see immediately whether a cheaper tier would save enough to be worth testing.

Runs entirely in your browser. Prices come from the same dated, source-linked table the rest of this site uses.

How to use the result

The spread is the point. Between the cheapest and most expensive model in that table there is usually a 50–100× difference for identical volumes. No prompt optimisation comes close to that, which is why model selection is the first cost decision, not the last.

Try the cache slider. Set it to 80% and watch what happens. For any workload with a stable prompt prefix — a fixed system prompt, few-shot examples, a retrieved document set — prompt caching is typically worth more than dropping a model tier, and it costs you no quality at all.

Then test, don't assume. A cheaper model is only cheaper if it does the job. Run your actual workload on the two or three cheapest candidates and measure quality before you migrate.

What this doesn't include

• Reasoning tokens. Reasoning models generate billed output you never see, so their real output volume is higher than what you'd estimate from the visible response.
• Batch discounts. Typically ~50% off both rates for work nobody is waiting on. If your workload is asynchronous, halve the numbers.
• Credits, free tiers and negotiated discounts, which are applied at billing time.
• Cache writes. The slider models cache _reads_. Writes cost 1.25–2× base input and pay back after one or two reads — see any model page for its exact break-even.
• Tokenizer differences. The same text is a different number of tokens on each provider, so cross-provider comparisons carry a margin of error. Measure on your real inputs.

For what you are actually being charged rather than what a calculator estimates, read the provider billing API — how to track spend across providers.