Token counter for ChatGPT, Claude and Gemini
Paste any text to see how many tokens it is on each AI, and what it costs to send or receive on every current model. It runs in your browser, so your text stays private.
What this text costs on each model
At today's pay-per-use list prices. "As a prompt" is the cost to send it; "as an answer" is the cost if the AI wrote it, which is usually about five times more.
| Model | Level | Tokens | As a prompt | As an answer | 1,000 prompts |
|---|---|---|---|---|---|
| GPT-6 LunaOpenAI | Light | ≈ 0 | $0 | $0 | $0 |
| GPT-6.1 TerraOpenAI | Everyday | ≈ 0 | $0 | $0 | $0 |
| GPT-6.1 SolOpenAI | Hard reasoning | ≈ 0 | $0 | $0 | $0 |
| GPT-6 AstraOpenAI | Frontier | ≈ 0 | $0 | $0 | $0 |
| Gemini 3.1 Flash-LiteGoogle | Light | ≈ 0 | $0 | $0 | $0 |
| Gemini 3.8 FlashGoogle | Everyday | ≈ 0 | $0 | $0 | $0 |
| Gemini 4 ArgonGoogle | Hard reasoning | ≈ 0 | $0 | $0 | $0 |
| Claude Haiku 5.5Anthropic | Light | ≈ 0 | $0 | $0 | $0 |
| Claude Sonnet 5.5Anthropic | Everyday | ≈ 0 | $0 | $0 | $0 |
| Claude Opus 5.5Anthropic | Hard reasoning | ≈ 0 | $0 | $0 | $0 |
| Claude Fable 5.1Anthropic | Frontier | ≈ 0 | $0 | $0 | $0 |
Want it shorter? The prompt optimizer rewrites a prompt to use fewer tokens without losing what you asked for, and the AI cost calculator turns these numbers into a monthly bill.
How this token counter works
AI companies don't charge by the word. They split text into tokens, pieces that are often a whole short word, part of a longer one, or a punctuation mark, and bill per million tokens. This counter runs OpenAI's open-source tokenizer (o200k_base) directly in your browser for an exact ChatGPT count, then estimates Gemini and Claude from it.
- ChatGPT (OpenAI): exact, using the same tokenizer as OpenAI's current models.
- Gemini (Google): estimated at the same count. Google's tokenizer counts English at about the same rate.
- Claude (Anthropic): estimated at about 35% more. Anthropic says current Claude models use about 30% more tokens than its older ones, and independent tests measured about 40% more than OpenAI for English.
Costs use each company's published pay-per-use prices, and they switch to announced new prices on the day those take effect. See today's AI prices for the full list.
Why the same text has different counts
Each company trained its own tokenizer on different text, so they split words differently. That matters because prices are quoted per token: two models with the same price per million tokens can still cost different amounts for the same job. Our guide to how AI tokens work has measured examples in English and ten other languages, and the hidden tokens that never show up in a chat window.
Turn token counts into a budget
One prompt costs a fraction of a cent. Thousands a day add up. The AI cost calculator turns prompt and answer lengths into a monthly bill on every major model, and our guide to cutting AI costs covers what to change first.
Frequently asked questions
How many tokens is a word?
About 1.3 tokens per English word on ChatGPT and Gemini, so 100 tokens is roughly 75 words. Current Claude models count roughly 35% more tokens for the same text. Code, numbers and most other languages use more tokens per word than plain English.
Is this token counter accurate?
The ChatGPT count is exact for OpenAI's current o200k_base tokenizer, run in your browser. Gemini and Claude counts are estimates: Gemini counts English at about the same rate as OpenAI, and current Claude models about 35% higher. For exact Claude or Gemini counts, use Anthropic's free token counting endpoint or Google's countTokens.
Why does Claude count more tokens than ChatGPT?
Each company splits text with its own tokenizer. Anthropic says its Claude 4.7 and later models produce about 30% more tokens than its older ones for the same text, and independent tests measured about 40% more than OpenAI's count for English prose. Since prices are per token, the same job can cost more on Claude at the same list price.
Does this tool store or send my text?
No. The counter downloads OpenAI's open-source tokenizer once and runs it on your device. Your text isn't sent to PromptSaver or anyone else.
Do prompts and answers cost the same per token?
No. Every major AI company charges more for tokens the model writes than for tokens you send, usually about five times more. That's why the table shows both: the cost to send this text as a prompt, and the cost if an AI wrote it as an answer.
How do I use fewer tokens?
Ask for the length you need, paste only the part of a document that matters, start a new chat when the topic changes, and use a smaller model for simple jobs. PromptSaver's optimizer rewrites a prompt to be shorter and clearer for you.