ChatGPT Token Counter
Estimate tokens before you send your prompt to GPT-4, GPT-4o, GPT-5, or any GPT-class model. See where your prompt sits against the context window and what percentage of your budget it uses.
Keyword density
Top 8Add a prompt to see your top keywords.
How ChatGPT counts tokens
ChatGPT uses OpenAI’s cl100k_base and o200k_base tokenizers, depending on the model. Common English words take 1 token. Rare words split into 2–4 tokens. Each punctuation mark and emoji usually counts as 1 token. Chinese, Japanese, and Korean text typically take 1–2 tokens per character.
This tool runs a calibrated heuristic that approximates these tokenizers within ~10%. The estimate is fast and runs entirely in your browser. For exact token counts (billing accuracy), use OpenAI’s official tokenizer or the tiktoken Python library.
ChatGPT context windows
| Model | Input window | Output max |
|---|---|---|
| GPT-3.5 Turbo | 16K tokens | 4K tokens |
| GPT-4 | 8K–32K tokens | 4K tokens |
| GPT-4 Turbo / 4o | 128K tokens | 4K–16K tokens |
| GPT-5 | 256K tokens | 32K tokens |
A prompt that fits in 80% of the context window leaves room for the model’s reply. Pushing past 90% almost always truncates output.