How much of the window does your prompt use? See it before you send it.

Paste a prompt, transcript or document — get the estimated tokens, the share of the model's context window, the cost, and the room left for the reply.

  • Nothing pasted ever leaves your device
  • 9 real models, verified windows
  • Warnings at 50%, 80%, 95%
  • Free, no account

Visualize context window usage

Paste your prompt, transcript or document

0 characters · 0 words · 0 lines Token count is an estimate (±~10% on prose)

Nothing pasted yet — try the example, or paste any text. Results update as you type.

Model & allocation

Tokens held back for the model's reply. A long answer needs a bigger reserve.

If your app adds a system prompt, enter its token count here.

Tokens from earlier turns, re-sent with every new message.

Used for the "turns left" estimate — a typical user message plus reply is a few hundred tokens.

Privacy you can verify, not just trust

Your pasted text is counted by JavaScript running on your own device — it is never sent anywhere. The only requests to other sites are the site's own analytics, which count page views. Don't take our word for it:

Third-party requests since load: 0 (site analytics only)
  1. Press F12 to open developer tools, then the Network tab.
  2. Paste any text above. Watch: no request carries your text anywhere — the only third-party requests are the site's own analytics.
  3. Extra credit: disconnect your internet after this page loads — counting still works.

How to check context window usage before you send

Every model has a fixed context window — the maximum number of tokens it handles in one request. The system prompt, the conversation history, any retrieved documents, your new message and the reply all share that one budget. When the total passes the limit, the API rejects the request or your app silently drops older history. Checking first is cheaper than debugging a truncated reply.

This visualizer does the check in your browser:

  1. Paste the text — a prompt, a chat transcript, a document you plan to attach. The tool estimates its tokens with a byte-pair-style heuristic and shows the count live as you type.
  2. Pick the model — nine real models with their published windows, or a custom window for any model not listed. The usage bar shows your text, system prompt, history and output reserve as separate segments.
  3. Read the verdict — the share of the window used, the tokens remaining, the estimated cost per request, and how many more turns fit. Warnings appear at 50%, 80% and 95%.
  4. Compare models — the table shows the same text against every listed model, so you can see at a glance which window actually fits it and what each one would cost.

Why a prompt costs more the longer you chat

Each new turn re-sends the entire conversation history as input tokens. A chat that starts at 500 input tokens can reach 5,000 after a dozen turns — and the input bill grows with it, even though you only typed a short message. Long system prompts and retrieved documents add the same way. The "turns left" figure is the practical version of this: how many more messages of a typical size fit before the window runs out.

Honest limitations

The token count is an estimate

Real token counts depend on the model's own tokenizer. This tool's heuristic lands within about 10% on plain English prose and reads a little high on code and JSON, where punctuation is dense. Use the provider's tokenizer for billing.

Prices are indicative list prices

Model prices are the providers' published list prices as of November 2026. Long-prompt surcharges are not applied — Gemini 2.5 Pro, for example, bills prompts over 200,000 tokens at a higher rate. Check the provider's pricing page before budgeting.

Model windows change

Providers raise context windows and launch new models regularly. The nine listed models were verified against published figures in November 2026; if a model you use is missing or its window has grown, use the Custom option and type the published number.

Chat apps add hidden tokens

Chat interfaces wrap your text in formatting, role markers and tool definitions that also consume tokens — typically a few hundred per turn. Treat the estimate as the floor, not the ceiling, of what a real request uses.

Frequently asked questions

How do I check how much of a context window my prompt uses?

Paste the prompt, chat transcript or document into the box, pick the model, and the visualizer estimates the token count and shows the share of that model's context window the text fills, the tokens remaining, and warnings at 50, 80 and 95 percent. Add your system prompt and conversation history tokens, and a reserve for the reply, to see the full allocation instead of the pasted text alone.

How accurate is the token estimate?

It is an estimate, not an exact count. The tool pre-tokenizes text the way byte-pair encoders do — short words are one token, long words split roughly every four characters, numbers split every three digits — which lands within about 10 percent of a real tokenizer on plain English prose. Code and JSON read a little high because punctuation is dense. For billing, always use the provider's own tokenizer; this estimate is for planning.

What is a context window, and why does it matter?

A context window is the maximum number of tokens a model handles in one request — the system prompt, the conversation history, retrieved documents, the new message and the reply all share it. When the total exceeds the window, the API rejects the request or the app has to drop older history. Knowing the share used before you send saves failed calls, truncated replies and surprise bills.

Why does my prompt cost more as the conversation grows?

Every new turn re-sends the whole conversation history as input, so the input token count — and the input bill — grows with every message. A chat that starts at 500 input tokens can reach 5,000 after a dozen turns. The visualizer's turns-left figure shows how many more turns of a given size fit before the window runs out.

What should I do when my prompt is nearly full?

Shorten the prompt first: cut repeated instructions, trim retrieved documents to the relevant passages, and summarise older history instead of re-sending it. Reserve fewer tokens for the reply if the answer can be short. If the content genuinely needs the space, move to a model with a larger window — the comparison table shows the same text against every listed model.

Are the model prices on this page exact?

No — they are indicative list prices per million tokens, published by the providers as of November 2026. Long-prompt surcharges are not applied: for example, Google bills Gemini 2.5 Pro prompts over 200,000 tokens at a higher rate. Open-weight models carry no price because the cost depends on the host. Treat the cost figure as a planning estimate and check the provider's pricing page before budgeting.

Is my pasted text uploaded anywhere?

No. All counting happens in JavaScript on your own device — the pasted text never leaves it. The only requests to other sites are the site's own analytics, which count page views. You can verify this in your browser's Network tab: paste text and watch — no request carries it anywhere.