AI Token Counter

Paste a prompt or document and see its estimated token count — and how much of each common context window it fills. The number your AI bill and context limits actually run on.

Free vs paid — when to upgrade

What this free tool is great for: a quick, one-off job with no signup — it runs entirely in your browser, so nothing leaves your device and there's nothing to manage.

Its honest limit: it's a one-off calculation in your browser — it doesn't save your scenarios, update as your real numbers change, or connect to your live accounts, so you re-enter the figures every time and can't watch how they move.

Where Dify does more: Counting tokens is planning; building is next. Dify is the open LLM-ops platform where prompts become real AI apps and agents — with retrieval, deployment and per-user token monitoring built in.

Try Dify →

Affiliate link — we may earn a commission if you sign up, at no cost to you. It never changes our honest take.

Tokens: the currency your AI actually charges in

Every large language model reads and writes in tokens — chunks of text that are usually smaller than a word and larger than a character. Your API bill is priced per token. Your context window is measured in tokens. Your prompt gets truncated at a token limit, not a word limit. Yet almost everyone reasons in words, which is why AI costs and cut-off prompts keep surprising people. This counter gives you the missing number: paste any text and see its estimated token count, plus how much of the common context-window sizes it consumes — counted entirely in your browser, with an honest label on the word "estimated."

What a token actually is

Tokenizers split text into pieces drawn from a fixed vocabulary learned from massive corpora. Common words ("the", "and", "founder") are usually single tokens; rarer words get split into sub-pieces ("tokenization" might become "token" + "ization"); punctuation, spaces and capitalisation all influence the split. That's why token counts feel unintuitive: "ChatGPT is great" and "Antidisestablishmentarianism!" have wildly different token-per-word ratios. For ordinary English prose the averages are stable enough to estimate — roughly four characters or three-quarters of a word per token — and those two rules of thumb, blended, are what this tool uses.

Why an estimate is honest — and when it drifts

Exact token counts require running the specific model's tokenizer, and every family (GPT, Claude, Llama, Gemini) tokenizes slightly differently — the same paragraph can differ by ten percent between them. For planning purposes that rarely matters: budgeting, context-fit checks and cost estimates all survive ±15% noise. But know when the estimate drifts further: code tokenizes heavier than prose (symbols and indentation fragment badly), non-English languages often cost dramatically more tokens per word — some scripts multiply counts several-fold — and text stuffed with URLs, IDs or JSON skews high. If you're doing precision work at scale, use the provider's own tokenizer; for everything else, this estimate is the right tool.

The context window: your token budget per request

A model's context window is the total tokens it can hold at once — your system prompt, the conversation history, the retrieved documents, the user's question and the answer it writes, all sharing one budget. The fit-bars above show your text against common window tiers, from small 8k models to the million-token giants. The practical insight most people miss: the window fills faster than you think, because everything counts. A 200-page document, a long chat history, a bloated system prompt — they all crowd out room for the answer. When a model "forgets" the start of a long conversation, it didn't forget; the early tokens fell out of the window.

Tokens are money: budgeting prompts like a CFO

Because pricing is per token — with output tokens typically costing several times input tokens — token counts translate directly to unit economics. A system prompt of 2,000 tokens sounds harmless until it rides along on every one of a million requests. The high-leverage moves, in order: trim the system prompt (it's paid on every call); summarise or truncate conversation history instead of resending it whole; retrieve only the relevant chunks of documents rather than stuffing entire files; and cap output length, since output is the expensive direction. Count before you ship — our AI Cost Calculator turns the token counts from this tool into an actual monthly bill projection.

Counting for writers and prompt-crafters

Token awareness isn't just an engineering concern. If you write prompts for a living — or paste long documents into chat assistants — the counter answers everyday questions: will this whole report fit, how much can I paste before the model loses the beginning, why did my carefully crafted instruction get cut. A useful habit for long tasks: check the token count of your material first, and if it brushes against the window you're using, split by sections with a running summary instead of trusting one giant paste. The models handle a well-structured 30k tokens far better than a truncated 130k.

Why counts differ between tools

Paste the same text into three token counters and you may get three numbers. That's not sloppiness — they're either using different tokenizers (each accurate for its own model) or different estimation rules. This tool deliberately shows a blended estimate rather than impersonating any specific model's tokenizer, and says so — an honest ~1,050 beats a false-precision 1,047 that's only true for one model family. Treat cross-tool differences under fifteen percent as agreement. When someone quotes you an exact token count without naming the tokenizer, they've told you less than this estimate does.

A budgeting workflow that sticks

The teams that keep AI costs boring all converge on the same loop. First, count the fixed overhead: paste your system prompt and any always-included examples here and treat that number as the per-request tax. Second, estimate the variable part — a typical user message plus retrieved context — and set the output cap deliberately rather than accepting the default. Third, multiply through your expected volume in our AI Cost Calculator to see the monthly bill before a line of code ships. Finally, re-count whenever the prompt changes: prompts only ever grow, a phenomenon veterans call prompt bloat, and a monthly five-minute audit of what's actually riding along on every request routinely finds a third of the tokens doing nothing. Counting is the cheapest optimisation in AI — it just has to actually happen.

Counting tokens is step one — where Dify does more

A counter tells you what a prompt costs and whether it fits; the harder work is building the thing around the prompt — the chains, the knowledge base with retrieval, the agent logic, the deployment, the observability that shows token spend per user in production. That's where Dify does more: an open LLM-ops platform where you visually build AI apps and agents, plug in your documents, and monitor exactly the token flows you estimated here — turning prompt-craft into shipped product without wiring the plumbing yourself. Count and budget here; when the prompt needs to become an app, build it on rails instead of glue code.

Frequently asked questions

More free calculators

Browse all free tools →

Embed this tool on your site

Blogger, teacher or toolmaker? Put this calculator on your own page — free forever, no strings. Copy the snippet below (the credit link is appreciated and keeps the tool free):

This tool is free and runs entirely in your browser. The link above is an affiliate link: we may earn a commission if you sign up, at no extra cost to you, and it never changes our honest take.