Count tokens for GPT, Claude & Gemini — before you hit send.
Paste any prompt, document, or codebase snippet and get exact token counts for GPT-4o, GPT-4, GPT-3.5, and o1 — plus estimates for Claude and Gemini. Check context-window fit and estimate cost, all without your text ever leaving this page.
🎯
Exact OpenAI counts
Runs the real cl100k_base and o200k_base encodings locally — same tokenizer OpenAI uses.
📏
Context-window fit
See at a glance whether your text fits each model's context window before you run out of room.
💰
Cost estimator
Editable per-model pricing turns your token count into an instant cost estimate.
Your text
Files are read locally with the File API and never leave your device.
Token counts
GPT-4o · o1 · o1-mini
0
tokens (o200k_base, exact)
GPT-4 · GPT-3.5
0
tokens (cl100k_base, exact)
Claude
0
tokens (≈4 chars, estimate)
Gemini
0
tokens (≈4 chars, estimate)
0
characters
0
words
0
lines
Context window fit & cost
Prices are editable — update them to match your plan.
| Model | Tokens | Context window | Fit | Input $ / 1M | Est. input cost |
|---|
How it works
Three steps from paste to token count — all in your browser.
1
Paste or drop text
Type, paste, or drag in a text file. The page reads it locally with the File API — nothing leaves your device.
2
Get exact counts
OpenAI's cl100k_base and o200k_base encodings run in your browser for exact GPT token counts, plus heuristic estimates for Claude and Gemini.
3
Check fit & cost
Context-window fit bars show whether your text fits each model, and the editable price table estimates your input cost per request.
Where does your prompt go? Nowhere.
Many online token counters paste your text into a server to count it — your prompts, code, and drafts pass through someone else's infrastructure. AI Token Counter works differently.
Upload-based token counters
- ✗ Your prompt or document is sent to a remote server for counting
- ✗ Drafts, proprietary code, and confidential text pass through third-party infrastructure
- ✗ Accounts, ads, or request limits gate the features
- ✗ No guarantee the text isn't logged or analyzed
AI Token Counter — 100% client-side
- ✓ Tokenization runs in your browser with OpenAI's real encodings — nothing uploaded
- ✓ No account, no signup, no tracking, no telemetry
- ✓ Safe for proprietary code, unpublished drafts, and confidential prompts
- ✓ Works offline after the page loads; text vanishes when you close the tab
Who is this for?
🐛
Developers
Check API costs and context fit before building prompts or shipping calls to OpenAI, Anthropic, or Gemini endpoints.
✍️
Writers
Verify drafts fit model limits and estimate cost before pasting unpublished work into an AI tool.
🔐
Privacy-conscious users
Count tokens on confidential material without uploading it — ideal for NDA-covered code and client data.
🧑🏫
Students & educators
Learn how tokenization and context windows work while budgeting AI usage for projects and assignments.
Frequently Asked Questions
Everything you need to know about counting tokens with AI Token Counter.
What is a token in AI models? ▾
A token is the smallest unit of text that a language model reads or writes. Roughly 100 tokens equal about 75 English words, but tokenization varies by language and content — code, punctuation, and emoji all consume tokens differently. Models bill usage, enforce context-window limits, and price API calls by token count.
How accurate is the AI Token Counter? ▾
For OpenAI models the count is exact: the tool runs the same cl100k_base encoding used by GPT-4 and GPT-3.5 Turbo, and the o200k_base encoding used by GPT-4o and the o1 family, directly in your browser. For Claude and Gemini the count is a close heuristic estimate (about 4 characters per token), because Anthropic and Google have not published their full tokenizers for client-side use.
Does it work for Claude and Gemini, or only GPT models? ▾
Both. GPT-4o, GPT-4, GPT-3.5 Turbo, o1, and o1-mini get exact counts from OpenAI's official encodings. Claude (Anthropic) and Gemini (Google) get a clearly labeled heuristic estimate, since their tokenizers are not publicly available for browsers. The context-window fit and cost rows for those models use the same estimate.
What is a context window and why does it matter? ▾
A context window is the maximum number of tokens a model can process in one request — your input plus its output. GPT-4o supports 128,000 tokens, o1 supports 200,000, and Gemini 1.5 Pro supports up to 2,000,000. If your text exceeds the window, it must be truncated or split. The tool shows a fit bar for each model so you can see at a glance whether your text fits.
Does the AI Token Counter upload my text anywhere? ▾
No. Everything runs in your browser using client-side JavaScript. Your text is never sent to any server, so it is safe to paste proprietary code, unpublished drafts, or confidential prompts. There is no account, no tracking, and the page keeps no copy of your text after you close it.
How do I use the cost estimator? ▾
Each model row has editable price fields for input and output per 1 million tokens, pre-filled with current list prices. As you edit the price fields or change your text, the estimated input cost updates instantly. Adjust the numbers to match your plan's pricing — batch discounts and cached-input rates vary by provider.
Related Guides
Go deeper into tokens, context, and AI agent workflows.
What Are AI Agent Skills? A Guide to Reusable AI Workflows
Learn what AI agent skills are, why they matter, and how to structure reusable instruction documents that coding agents can load and follow across sessions.
Read the guide → Prompt GuideHow to Write Better AI Image Prompts
Structure prompt elements for clearer results and learn how token budgets shape what your prompt can say — from subject to style to composition.
Read the guide →Related Prescosoft Tools
Complement your AI workflow with these free, privacy-first browser tools.