How tokens work at AI2BIHub

Tokens are the unit of AI processing. Here's a live estimator, the per-call breakdown, and answers to the questions we get the most.

Token estimator

Paste any text below to see roughly how many tokens it represents. Updates as you type — nothing leaves your browser.

Characters
131
Words
23
≈ Tokens
33

Roughly 4 characters make one token in English. This counts the text itself — what you are charged also depends on how much we had to read to answer and how careful an answer you asked for, both covered below.

Where the tokens go in one question

Asking a question costs three things — what you typed, the part of your files we had to read to answer it, and the answer itself. The same question can cost different amounts on different days, because how much we need to read depends on what you asked and which files are selected.

Your question
What you typed into the chat input.
~50–200
Reading your files
The parts of your files we had to read to answer. A broad question across big files means reading more; a specific question means reading less.
~1,500–4,000
The answer
The answer you get back. A table or a long list costs more than a one-line reply.
~300–800
Typical total per chat
Why most questions land in the 2,000–5,000 token range.
~2,000–5,000

The numbers above are typical, not a quote. Every question you ask is counted exactly and listed on your usage page.

Two other things that use tokens

Questions are the main one, but not the only one. Both of these are listed on your usage page like everything else.

Reading a document you upload
A file has to be read before anything can be asked about it. Charged by the page, so a fifty-page contract costs more to take in than a one-page invoice. Scanned and photographed pages cost a little more than typed ones, because we have to look at the page as a picture to pull the figures off it. Either way you pay this once per file, not every time you ask about it.
8,000 / page
Using the rest of the product
Opening a report, building a dashboard, running a breakdown. No AI is involved, but the work is real. These are charged at a small flat rate, and repeated clicks within a few seconds are only counted once.
~50

Why some answers cost more than others

Not all questions take the same amount of work. Quick ones are cheap and cover most of what anyone does. The slower, more careful kind cost us several times more to run, and we pass that difference on rather than charging everyone the top rate all day. The product always tells you when it is about to use the expensive kind.

Everyday questions
Summaries, lookups, straightforward answers. This is most of what anyone does.
standard rate
Harder reasoning
Multi-step analysis, comparisons across many files, anything that needs care rather than speed.
5x standard
The most careful kind
Kept for the hardest questions, where a better answer is worth paying for.
25x standard
Re-reading something we already read
When the same file or conversation is used again, we reuse what was read the first time instead of paying for it twice.
a tenth of standard

FAQ

What exactly is a token?+

A token is a small piece of text — about four characters in English, or roughly three quarters of a word. "Hello, world!" is about four of them. It is simply the unit everything is measured in, so a long question and a short one can be compared fairly.

Why does the same question cost different amounts?+

Two reasons. How much we have to read changes — the same question over a 5-page file means reading far less than over a 500-page one. And harder questions are answered more carefully, which costs several times more per question; the product tells you when it is about to do that. Narrower questions over fewer files are always cheaper.

Can I make it cost less?+

Yes. Ask narrower questions, pick only the files you actually need rather than everything, and ask for a summary instead of a long itemised list when the summary would do. The biggest saving is usually on the files themselves — upload the twenty pages that matter rather than the whole two-hundred-page bundle, and only re-upload something if it has actually changed.

How far does a pack actually go?+

Asking questions: 50,000 tokens is roughly three questions a week for a month. 200,000 is a dozen or so a week. A team running daily reports gets through 1,000,000 or more. Uploading files is the bigger number for most people — reading in a fifty-page document costs about the same as a hundred questions, so if your month has a lot of paperwork in it, size your pack on the paperwork rather than the questions.

Is there a monthly fee?+

You get 70,000 tokens free when you sign up, with no card and no trial clock. After that you buy tokens in one-off packs whenever you want more. There is no monthly fee and no monthly reset — whatever you have left stays yours.

What happens when my balance hits zero?+

You'll be asked to top up the next time you ask a question. Nothing is deleted and you are never charged automatically — your files, saved queries and history stay exactly as they were until you decide to add more.

Do you train AI on my data?+

No. We read your data to answer your question and nothing else. Neither your files nor your questions are used to train anything. The full list of who processes your data is on the legal pages.

Is the token count auditable?+

Yes. Every single charge — questions, files read in, and ordinary use of the product — is listed on your usage page with the time, what it was for, and what it cost. Download the lot as a spreadsheet whenever you want.

Ready to spend some tokens?

See plan tiers and what each one includes.