How to use TokenCost

Estimate your LLM bill in three steps.

TokenCost turns confusing per-token pricing into a monthly dollar figure. Here’s how to go from "which model?" to "what will it cost me?" — and how to read every number along the way.

Browse rates on Pricing

Every model is normalized to USD per 1 million tokens, so you can read across providers on the same scale. Filter by provider and sort by input price, output price, or context window to shortlist candidates.

Line up candidates on Compare

Pick up to four models to see input, output, cached-input, and context side by side. The cheapest value in each row is highlighted so trade-offs are obvious at a glance.

Project your bill on Calculator

Enter your expected traffic. The calculator converts it into monthly token volume and prices it against every model, so you see real monthly cost — not just a per-token rate.

01 — The inputs

What each input means

Rough estimates are fine — the goal is an order-of-magnitude picture.

Monthly active users

How many distinct people use your app in a month. If unsure, use your best launch-target number.

Prompts / user / day

How many times an average user triggers a model call each day. One chat message, one summary, one search — each is a prompt.

Avg input tokens

The size of what you send per prompt — system instructions, context, and the user message combined. Rule of thumb: ~750 words ≈ 1,000 tokens.

Avg output tokens

The size of what the model returns per prompt. Output is usually priced far higher than input, so this drives cost more than you would expect.

Prompt cache hit rate

The share of your input that repeats across calls (a fixed system prompt, a shared document). Cached input is billed at roughly 10% of the normal rate. Leave at 0 if unsure.

02 — The math

How the number is calculated

prompts/mo = users × prompts/day × 30

input tokens = prompts/mo × avg input

output tokens = prompts/mo × avg output

cache factor = 1 − 0.9 × cache hit rate

monthly cost = input ÷ 1M × input price × cache factor

+ output ÷ 1M × output price

Every model runs through this same formula with its own input and output prices, then they’re ranked cheapest to most expensive.

03 — The output

Reading the results

Tier cards

A representative Budget, Standard, and Premium model with its monthly total and a per-user figure — a fast sense of the cheap / balanced / top-quality options.

The ranked table

All models sorted low to high. The Relative bar shows each model’s cost as a share of the most expensive one, and the cheapest option is flagged.

Per-user cost

Monthly total divided by active users. Compare this against what you charge per user to sanity-check your margins.

Reference

Key terms

token

A chunk of text a model reads or writes — roughly ¾ of a word. Billing is per token.

input vs output

Input is what you send; output is what the model generates. Output almost always costs more.

cached input

Repeated input the provider stores and re-serves at a steep discount (~10% of input).

context window

The maximum tokens a model can consider at once — input plus output combined.

Want three fully worked scenarios (chatbot, RAG, coding agent) instead of the abstract formula? See How to Estimate LLM Costs Before You Build.

Try the calculator →