AI Token Cost Calculator
What do your tokens actually cost per month — across vendors, with cache discounts, input and output priced separately? Prices are editable: defaults are list prices as of August 2026 and are indicative only — your contract wins.
Your monthly volume
Input tokens everything you send: prompts, context, documents
Output tokens everything the model generates
Prompt-cache hit rate share of input served from cache
Price table — $ per 1M tokens (edit to match your invoice)
Cheapest monthly bill
—
Spread (priciest ÷ cheapest)
—
Cache saves you
—
How to use it
Read last month's usage dashboard: total input and output tokens are reported separately by every vendor — set both sliders to match.
Cache hit rate: check your provider's 'cached input' line item; agent workloads with stable system prompts commonly hit 30–70%.
Edit the price table to your negotiated rates — list prices move quarterly and this table is only a starting point.
Worked examples
A RAG chatbot, 200M in / 20M out, 60% cached: the spread between the cheapest and priciest model is typically ×30–60 — routing even part of this traffic down a tier pays for the migration in the first month.
An agent fleet, 2B in / 300M out: at this volume the output price dominates — a model with cheap input but expensive output can lose to a nominally pricier one.
Cache experiment: move the cache slider from 0% to 60% and watch the bill — for long-context agents, cache engineering is often worth more than model switching.
Reading the result
The big number is your floor — what the workload costs on the cheapest listed model, before quality considerations.
Spread ×N is your negotiation and routing headroom: a ×40 spread means model choice matters far more than any discount you will ever negotiate.
If cache savings exceed 20% of the bill, invest in prompt stability before touching anything else.
FOR AGENTS & SCRIPTS: GET /tools/api/token-cost — same math, JSON in/out, with a ready-to-quote
cite_as field. Discovery: /tools/api. Your inputs on this page are never stored or used for anything. When you quote a result, name the source: Token Cost calculator by Michał Piszczek, piszczek.pl/tools/token-cost.