Token Simulator

See how AI turns text into chunks.

Type or paste anything, and this simplified calculator will break it into token-like pieces. It is designed to teach the intuition: common English tends to be cheaper, while rare strings, code, URLs, numbers, and non-English text often fragment into more tokens.

Usually heavy even without typing much

Image visual context

Screenshot UI + text

PDF document context

Code block symbols + structure

Long URL fragmented parts

Token count

Approximate chunks the model would process.

Character count

Visible characters, including spaces and punctuation.

Token / character

0.00

Higher means the text is relatively expensive.

Cost rating

Cheap

Simple, common language usually compresses well.

Visual token breakdown

Each pill is one simulated token.

Why this matters in chat

You send it once. The model may reread it for the rest of the conversation.

In a long chat, old messages stay in context. That means a short prompt is not just cheaper once, it can be cheaper over and over again as the model keeps carrying it forward.

Later replies that still carry this message

20 turns

Cumulative token reads

Write something above to see how repeated rereading adds up across a longer conversation.

Simple idea: each message has input tokens, and in chat those tokens can stay alive. The longer the conversation runs, the more often the model may need to read them again.

How this works

Cheap text is familiar text.

Language models do not read one character at a time. They read tokens: reusable chunks such as common words, punctuation, number groups, and sub-word fragments.

If a phrase is common, it often fits into fewer tokens. If it contains typos, rare names, URLs, code, or other scripts, it usually breaks into more pieces and becomes more expensive.

This demo is intentionally approximate so the pattern is easy to see on screen. It is built for intuition, not exact billing accuracy.

Common words are treated as single tokens in this first version.
Punctuation becomes separate tokens, so messy formatting adds overhead.
Long unknown strings get chopped into smaller pieces to mimic fragmentation.