Token to words converter
Words and tokens are not the same thing, and the conversion rate depends on the model. This page gives you the working numbers: the standard 1 token ≈ 0.75 English words ratio, a preset table for the word counts people actually plan around, and the real output cost of each count across every model we track.
Convert with the formula
Words → tokens
tokens = words ÷ 0.75
A 1,000-word article is about 1,333 tokens of input.
Tokens → words
words = tokens × 0.75
1,000 tokens is about 750 words of English prose.
Need exact counts for your own text? The interactive token ↔ words counter runs a real tokenizer in your browser — paste the text and compare the actual token count with the 0.75 estimate for GPT, Claude and Gemini.
Preset word counts → tokens
The estimates below use the 0.75 words-per-token ratio for plain English prose. Code, JSON, URLs and non-Latin scripts tokenize denser — expect 20–40% more tokens for those.
| Words | ≈ Tokens | Typical content |
|---|---|---|
| 250 | 333 | Short email or blog intro |
| 500 | 667 | Product page or news brief |
| 1,000 | 1,333 | Standard SEO article |
| 2,000 | 2,667 | In-depth guide section |
| 5,000 | 6,667 | Long-form report or whitepaper |
| 10,000 | 13,333 | Ebook chapter / full report |
What 1,000 words of output costs on each model
Output tokens are priced roughly 4–8× input tokens, so word count is the dominant cost driver for chat and generation workloads. Prices are USD per 1,333 output tokens (1,000 words), from our pricing table last verified 2026-10-20.
| Model | Vendor | Cost / 1,000 words out | Cost / 10,000 words out |
|---|---|---|---|
| GPT-5 nano | OpenAI | $0.0005 | $0.0053 |
| DeepSeek V3.2 (chat) | DeepSeek | $0.0006 | $0.0056 |
| GPT-4o mini | OpenAI | $0.0008 | $0.0080 |
| Llama 4 Maverick | Meta | $0.0011 | $0.011 |
| GPT-5 mini | OpenAI | $0.0027 | $0.027 |
| DeepSeek R1 (reasoner) | DeepSeek | $0.0029 | $0.029 |
| Gemini 2.5 Flash | $0.0033 | $0.033 | |
| Kimi K2 | Moonshot | $0.0033 | $0.033 |
| Claude Haiku 4.5 | Anthropic | $0.0067 | $0.067 |
| Mistral Large 2 | Mistral | $0.0080 | $0.080 |
| Qwen3 Max | Alibaba | $0.0080 | $0.080 |
| GPT-5 | OpenAI | $0.013 | $0.133 |
| GPT-4o | OpenAI | $0.013 | $0.133 |
| Gemini 2.5 Pro | $0.013 | $0.133 | |
| Claude Sonnet 4.5 | Anthropic | $0.020 | $0.200 |
| Grok 4 | xAI | $0.020 | $0.200 |
| Claude Opus 4.1 | Anthropic | $0.100 | $1.000 |
Why the ratio is not a rule
Tokenizers split text into subword units, so the words-per-token ratio depends on vocabulary, punctuation density and language. English prose sits around 0.75 words per token. Technical text with symbols, numbers and identifiers drops toward 0.5. Chinese, Japanese and Korean are much denser: one character often costs one to two tokens, so 1,000 characters can be 1,000–2,000 tokens rather than the ~500 a naive character-count estimate would suggest.
That is why the table above is an estimate and the browser tokenizer exists: for anything you are going to bill or budget, paste the real text and use the actual count. Then take that count to the prompt cost estimator to turn it into a monthly figure.
How to plan from word count
If you know the length of your typical answer, you already know most of your cost. A support bot that replies with 150 words spends about 200 output tokens per turn; a summarizer producing 500-word digests spends about 667. Multiply by your monthly request volume and compare models in the budget calculator to see which tier your volume actually supports.