Token counter
Paste text or code to estimate the token count across Claude, GPT and Gemini tokenizers. Everything runs in your browser.
- GPT (o200k / cl100k)44
- Claude47
- Gemini42
Estimate. The exact count depends on the model's tokenizer and the text; code and symbols tokenize denser than prose. What is a tokenizer?
How many tokens is your text?
In everyday English, expect roughly 0.75 words per token — about 1,300 tokens for 1,000 words. Code is markedly denser: braces, indentation and variable names produce a lot of tokens for very few characters. That's why a 200-line source file often costs more than a page of prose of the same byte size.
Why Claude, GPT and Gemini don't agree
Each model family ships its own tokenizer, trained on a different corpus. A word that was frequent in one model's training data becomes a single token there and two or three elsewhere. The gaps stay small on English prose but widen on code, non-English text and unusual identifiers. Treat the number here as a solid estimate, not an invoice.
What those tokens cost
APIs bill per million tokens, with output priced around five times input. In an agent session, though, the traffic is overwhelmingly inbound — files read, command output, and the whole conversation replayed on every turn. Input volume is what sets the bill, even at the cheaper rate.
Per-model rates are laid out in Claude Code pricing, and the levers for sending fewer of them in how to reduce token usage.
Frequently asked questions
- Is the count exact?
- It's an estimate. The exact token count depends on the model's BPE tokenizer and the precise text. Code and symbols produce more tokens per character than prose, which the tool accounts for.
- What's the difference between tokens and words?
- A token is a fragment of text — often a word, sometimes part of a word or a punctuation mark. In English, expect roughly 0.75 words per token; code is denser.
- Is my data sent anywhere?
- No. All counting happens in your browser. Nothing is uploaded or stored.
- Why does token count matter?
- LLM APIs bill per token and every model has a limited context window. Fewer tokens means a lower bill and more headroom before you saturate the context. Tokenade automatically trims your AI coding agent's tokens.
- How many tokens is 1,000 words?
- Roughly 1,300 tokens in everyday English. Code is much denser — closer to 1 token per 3 characters. Paste your actual text above rather than relying on the rule of thumb.
- Does this match what the provider actually bills?
- Very close, but not identical. Providers bill on the exact tokenizer of the model version you call, and add system prompt and message-formatting tokens that are not visible here. Use this to size and compare, not to reconcile an invoice to the cent.
- Why did my agent burn so many tokens when I only typed three lines?
- Because you are not billed for what you type. One agent turn carries the files it read, command output, MCP tool manifests, and the entire prior conversation — replayed every turn, because the model is stateless.
- How do I reduce my token count?
- Send the model less of what it never needed: targeted retrieval instead of whole-file reads, filtering on noisy command output, reading structure before full implementations, and loading tool manifests lazily. That is what Tokenade applies automatically.
Stop counting. Let Tokenade cut the bill.
Tokenade automatically trims your AI coding agent's tokens. Free up to ~20M tokens saved per month.
Start free