Token Counter
Estimate tokens and context usageRuns locally · nothing uploaded
Estimated tokens by model
Estimated from character classes, usually within ±15%. Images, files, tool definitions and system prompts also use context. The usage field returned by the API is the exact number. Text is counted in your browser and never sent to a model.
Context windows of common models
Click a row to estimate with that model's tokenizer and window.
| Model | Context window | This text uses |
|---|---|---|
| 400,000 | 0.013% | |
| 1,047,576 | 0.005% | |
| 128,000 | 0.039% | |
| 200,000 | 0.025% | |
| 1,000,000 | 0.006% | |
| 1,000,000 | 0.006% | |
| 200,000 | 0.028% | |
| 1,048,576 | 0.005% | |
| 1,048,576 | 0.005% | |
| 128,000 | 0.05% | |
| 32,768 | 0.156% | |
| 10,000,000 | 0.001% | |
| 128,000 | 0.039% |
Standard context windows as published by each provider (input and output combined); some models also have long-context or beta variants. Models change often — check the official docs (Claude data as of September 2026).
How to use
- 1Paste text or open a text file.
- 2Compare the estimates, or click a model in the context-window table below.
- 3Adjust the context window and reply budget, and optionally a price to estimate cost.
FAQ
How many tokens is a word?
In English, roughly 1.3 tokens per word or about 4 characters per token. Chinese and Japanese use around 0.6–1.1 tokens per character depending on the model.
Why doesn't it match the API exactly?
Every provider uses a different tokenizer. This tool estimates from character classes, usually within ±15%. The usage field returned by the API is exact.
Why reserve tokens for the reply?
The context window holds both input and output. If the input fills it, the model has no room left to answer and long replies get cut off.
Does it call a model API?
No. Every number is an estimate computed in your browser from character classes; nothing is sent to a model and nothing is billed. The context-window table is for reference — check the provider docs when models change.