Context Window Visualizer

How to Use the Context Window Visualizer

The Context Window Is a Budget, Not a Suggestion

A million is not one number: Claude 1,000,000, Gemini 1,048,576

Frequently Asked Questions

• What is a context window?+
  It's the maximum number of tokens a model can work with in one request. Anthropic counts everything in it — your system prompt, conversation history, pasted documents and the model's reply — and its API refuses an input that alone is over the window. Google states separate input and output limits for Gemini instead.
• Does the window include the model's response?+
  On Claude, yes: Anthropic defines the window as all the text the model can reference "including the response itself", so the reply is carved out of the same budget. Gemini lists a separate output limit (65,536 tokens) beside its 1,048,576-token input limit. Either way, leave headroom for the response you expect.
• Why are the GPT bars exact but Claude and Gemini estimates?+
  OpenAI publishes its tokenizer (tiktoken), so GPT counts are computed exactly in your browser — for the models tiktoken maps, which do not yet include GPT-6. Anthropic and Google don't release a browser tokenizer, so Claude bars use Anthropic's published characters-per-token figures and Gemini bars an estimate fitted to Google's own token counter — the same estimates as our Token Counter. xAI, DeepSeek and Mistral publish no figure we can use, so their bars borrow the GPT count. All of these are labelled "est."
• How accurate are the estimated bars?+
  They are planning figures, not counts. The Claude estimate follows Anthropic's own English figure (about 3.5 characters per token before Claude 4.7, about 30% more tokens on 4.7 and later); Anthropic gives no figure for Chinese, Japanese, Korean or Thai, so for those scripts it rests on OpenAI's rates. The Gemini estimate is set never to come out below Google's own counts on our test text. If an estimated bar sits near the top of a model's window, confirm with that provider's own token-counting endpoint before relying on it.
• My text is over a model's window — what now?+
  Either pick a model with a larger window, trim the input, or split it into pieces that each fit. Our Text Chunker can break long text into overlapping chunks sized to a token budget, which is the standard approach for RAG and long-document workflows.
• Is a bigger context window always better?+
  Not always. Anthropic says that as token count grows, accuracy and recall degrade ("context rot"), and a huge prompt costs more and runs slower. Fitting within the window is necessary but not sufficient — concise, well-structured context usually beats simply filling the window.
• Why isn't "1M tokens" the same for every model?+
  The "1M" figure is approximate marketing. The exact windows differ: Claude's 1M models are 1,000,000, Gemini's input limit is 1,048,576 (2²⁰) on every Gemini model listed here, and OpenAI's GPT-5.x are 1,050,000. This tool uses each model's own figure so the percentages are right, not a shared round number.
• Does my text get sent anywhere?+
  No. The tokenizer runs entirely in your browser from a locally-served library. Your text is never uploaded to any server or model; the only network request is your browser fetching the tokenizer file from our own domain.
• Why does the count appear a moment after I paste?+
  The exact tokenizer table loads on first use to keep the page fast. You'll briefly see a per-script estimate, then the exact GPT count snaps in. After that first load it's instant for the rest of your session.
• Is the tool free?+
  Completely free, with no account or sign-up, and no limit on use. It runs in your browser and collects no data.

Related Tools

Stored only in this browser — never sent to our servers.