← All tools

Token & Prompt Calculator

What will this prompt cost?

Paste a prompt and see what it costs to run, on every major model. Get an estimated token count, the cost per call and per month, and what caching and batching would save. Every rule we use is on the page.

  • About 3 minutes
  • List prices checked
  • Private to your browser

Step 1 · Your prompt

Paste the prompt you plan to run

Paste an instruction, a document or both. We loaded a short invoice-triage prompt as sample data. Every count updates as you type, and every count is an estimate.

The role, the task and the rules. Usually the same on every call.
Policies, vendor lists, a chart of accounts. Long, and often fixed.
Worked input and output pairs that show the model what good looks like.
Paste anything: a prompt, an email, a contract clause, a spreadsheet export.
Why split the prompt into parts
Most business prompts have a fixed part and a changing part. The instructions, policy and examples stay the same; the invoice or ticket changes every call. Providers can cache the fixed part and bill repeat reads at a fraction of the input price, but only if it comes first and is identical every time.

Step 2 · Answer and volume

Answer length and volume

Output tokens per call. A one-word label is under 20; a drafted memo is 1,000 or more.
Why the answer length matters
Output tokens cost about five times as much as input tokens on every model in our price list. Writing is slower, heavier work for the model than reading. Asking for named fields instead of prose is often the cheapest change you can make.
How many times this prompt runs: invoices, tickets or documents processed each month.
Why agents use more calls
A chatbot answers once. An agent works in a loop: plan, call a tool, read the result, decide again. Each step resends the instructions and everything gathered so far, so one task can be several calls that each grow. Multiply your calls per month by the steps per task. More in How much will AI agents cost?

Per call · estimated

Total tokens per call
0±15%
Input
0
Output
0

Prompt size

Share of tokens

Share of cost

    What this means

    Email me this

    Every model

    This prompt on every model, cheapest first

    Same prompt, same answer length, same volume, with your caching and batching settings. Price is one input: test quality on your own cases before you choose.

    Loading prices…

    Savings

    Caching and batching

      Cache hits are billed at each model's cached-input rate. Cache-write and storage fees are not modelled, and providers only cache a prefix above a minimum length, so treat small savings as an upper bound.

      How we count

      The estimating rule, applied to your text

      No tokenizer library: a plain rule you can check. Each provider splits text differently, so real counts vary by roughly ±15%.

      See the pieces

      Roughly where our rule cuts your text, for the first 600 characters. Alternating shades are separate estimated tokens; ↵ is a line break.

      Myth

      An agent costs about the same as one chatbot answer.

      Reality

      An agent calls the model several times per task and resends the growing context each time. Price the prompt here, then multiply by the steps. Read how agent costs add up.

      Field note

      In our experience the cheapest call is rarely the right goal on its own. Pick the smallest model that handles your hard cases, keep a person approving decisions that carry financial or customer risk, and cap automation around 80% of the task. A cheap model that routes the wrong invoice costs more than the tokens.

      Next step Budget the whole workload Add growth, several workloads and a monthly budget in the AI Budget Planner.

      Prices used

      List prices per million tokens

        This is an estimate from public list prices and a counting rule, not a bill. It does not include enterprise agreements, taxes, cache-write and storage fees, or tool and search fees. Long-prompt rates are applied to input and output where the price list gives them; the cached rate is left as listed. Check your contract and billing console before you commit a budget.

        Field guide

        Three lessons on tokens

        What a finance or operations leader needs to read an AI bill, and to ask the right questions of the team building the prompt.

        .01

        A token is a chunk, not a word

        Models read text in pieces. A common short word is usually one piece; a long or rare word is two or three. Numbers, punctuation and code break into more pieces, so a spreadsheet export costs more tokens than prose of the same length. Each provider cuts differently, which is why counts vary.

        Rule of thumbAbout 4 characters of English per token, or 750 words per 1,000 tokens.

        .02

        Writing costs more than reading

        The model reads your whole prompt in one pass, but writes its answer one token at a time. That is why output is priced several times higher than input. A short prompt that asks for a long essay can cost more than a long document that asks for three fields.

        Rule of thumbAsk for named fields or a fixed length. Prose is the most expensive format.

        .03

        Cache what repeats, batch what waits

        Caching stores the fixed start of a prompt so repeat reads bill at a fraction of the input price. It only works if that part comes first and is identical every call. Batching trades speed for price: send work that can wait up to a day and pay half.

        Rule of thumbPut fixed text first, changing text last. Batch anything no one is waiting on.

        Want a tool built for your team?

        We build custom assessments, skills and internal apps as part of our engagements. Tell us what you need.

        Book a discovery call contact@baxtermountain.ai