Token: the unit language models actually count
A token is the piece of text a language model reads or writes in a single step. It is often a whole short word, but long or unusual words are split into several tokens, and punctuation marks usually get their own. "Humanizer" might be two or three tokens; "the" is one. Models measure everything in tokens: how long your prompt can be, how much a response costs, how far back they can remember. Humanize 360 counts in plain words instead, and this page explains why the two are not the same.
Here is the practical rule of thumb for English prose: a word is a bit more than one token on average, so a 750-word essay is roughly a thousand tokens. That ratio drifts. Technical vocabulary, names, code and non-English text all use more tokens per word, because the model's vocabulary was built mostly from common English fragments. A tokeniser is the small program that does the splitting, and each model family has its own.
Why models work in tokens at all
A model needs a fixed list of symbols it knows. A list of every English word would be huge and still miss new words, names and typos. A list of letters is tiny but makes every prediction painfully slow. Tokens are the compromise: a vocabulary of tens of thousands of common fragments that can spell anything by combination. Predicting the next token is the whole job of a language model.
Where tokens show up for a writer
- Context limits: a model can only "see" a certain number of tokens at once, so very long documents get truncated or summarized.
- Pricing: API access to a model is billed per thousand tokens in and out.
- Truncated answers: a response that stops mid-sentence usually hit a token cap, not a content limit.
- Detectors: some detection methods score the probability of each token in sequence, which is where perplexity comes from.
Why Humanize 360 bills in words
Because you can count words and you cannot easily count tokens. One word in equals one credit, whether the word is "a" or "photosynthesis". A 5,000-word pack humanizes 5,000 words. That is possible because our engine is not a language model and does not tokenize the way one does; it works on sentences, words and punctuation directly. It also means a French name or a chemistry term does not quietly cost you three times as much.
A quick example
The sentence "Lagos traffic is unpredictable." is four words and a full stop. A typical tokeniser might produce five or six tokens: "Lagos" may be split, "unpredictable" almost certainly is, and the full stop stands alone. On Humanize 360 the same sentence costs four credits, and the Changes view shows edits at the word level, which is the level you actually read at.
Common questions
How many tokens are in a word?
For ordinary English, a little more than one token per word on average. Long words, names, numbers and code use more. Every model family tokenizes slightly differently, so treat any ratio as approximate.
Does Humanize 360 charge by token?
No. One word submitted is one credit, in every mode and at every strength. Word counts are shown before you run, so there are no surprises.
Are tokens the same as credits?
No. Tokens are a model's internal unit. Credits are our billing unit and equal words. The two are unrelated because our engine is not a language model.
Why does a chatbot cut off mid-sentence?
It reached its output token limit. Ask it to continue, or ask for a shorter answer. This is a limit of the model, not a sign that the text is finished.