What is a token in AI?
Large language models such as ChatGPT, Claude and Gemini do not read words; they read tokens, which are fragments of text produced by a tokeniser. A token is about four characters of English, so 100 tokens is roughly 75 words and a 1,000-word article is about 1,300–1,400 tokens. Punctuation, spaces and line breaks all count. This token calculator estimates the count from character length, which tracks the OpenAI and Anthropic tokenisers within about 10% for ordinary English prose.
Why Hindi, code and tables use more tokens
Tokenisers are trained mostly on English, so Devanagari, Tamil, Bengali and other Indian scripts are split into many small pieces: the same sentence in Hindi can cost two to four times the tokens of its English translation. Source code, JSON and markdown tables also tokenise badly because of symbols and whitespace. If your application serves Indian languages, budget for roughly three times the tokens you would estimate from an English draft.
Context limits by model
| Model family | Context window | Approx. words |
|---|---|---|
| GPT-4o / GPT-4.1 | 128,000 tokens | ≈ 96,000 |
| Claude | 200,000 tokens | ≈ 150,000 |
| Gemini 1.5 / 2.x | 1,000,000 tokens | ≈ 750,000 |
| Llama 3 (open) | 128,000 tokens | ≈ 96,000 |
The calculator shows a tick or cross for each family. Remember the context must hold your prompt, any documents you attach, the conversation so far and the model's reply, so leave 10–20% headroom.
How tokens turn into money
APIs charge per million tokens, with separate input and output prices; output is usually three to five times more expensive. Multiply the token count above by your model's price and by the number of requests you expect, or use the LLM API cost calculator linked on this page to get a monthly bill. Trimming system prompts, caching repeated context and capping response length are the three fastest ways to cut spend.
Frequently asked questions
How many tokens is 1,000 words?
About 1,300–1,400 tokens in English. Hindi or other Indian-language text can be 3,000–5,000 tokens for the same word count.
Is this count exact?
It is an estimate within roughly 10% for English. For billing-grade accuracy use the provider's tokeniser; for budgeting and context-limit checks this is sufficient.
Do ChatGPT and Claude count tokens the same way?
No, each has its own tokeniser, but counts for English are usually within 10–15% of each other.
Is my text uploaded?
No. The estimate is computed in your browser; nothing is sent to any server.