1 Million Tokens to Words: The Complete Conversion Guide
One million tokens is approximately 750,000 English words. That is the single most important number to remember when budgeting AI API costs. This guide explains the conversion, shows where the rule breaks down, and gives you a quick reference table for any token count.
The quick answer
1 million tokens = ~750,000 words. The standard rule of thumb is 1 token equals approximately 0.75 English words, or equivalently, 1 word equals about 1.33 tokens.
This ratio comes from how modern tokenizers (like OpenAI's tiktoken or Anthropic's tokenizer) split text. Common English words become single tokens, while longer or rarer words get broken into two or three sub-word pieces. Punctuation and whitespace also consume tokens.
Tokens to words: multiply by 0.75
Words to tokens: multiply by 1.33
Quick reference table
| Tokens | Words | Pages (~500 words) | Novels (~90K words) |
|---|---|---|---|
| 1,000 | 750 | 2 | 0.0 |
| 5,000 | 3,750 | 8 | 0.0 |
| 10,000 | 7,500 | 15 | 0.1 |
| 50,000 | 37,500 | 75 | 0.4 |
| 100,000 | 75,000 | 150 | 0.8 |
| 200,000 | 150,000 | 300 | 1.7 |
| 500,000 | 375,000 | 750 | 4.2 |
| 1,000,000 | 750,000 | 1,500 | 8.3 |
| 2,000,000 | 1,500,000 | 3,000 | 16.7 |
How tokenizers actually work
AI models do not read words — they read tokens. A tokenizer is an algorithm that splits text into pieces the model can process. Modern tokenizers use Byte-Pair Encoding (BPE), which learns the most common character sequences from a large corpus of text.
Here are some examples of how common words tokenize:
the→ 1 tokenhello→ 1 tokenunderstanding→ 1-2 tokensantidisestablishmentarianism→ 5-6 tokens你好(Chinese: "hello") → 2-3 tokens
When the 0.75 ratio breaks down
The 0.75 words-per-token ratio is an average for standard English prose. Several common scenarios produce different ratios:
- Source code: ~0.5 words per token. Brackets, semicolons, indentation, and camelCase identifiers inflate the token count.
- Technical writing: ~0.65 words per token. Jargon, abbreviations, and formulas tokenize less efficiently.
- Conversational text: ~0.80 words per token. Short, common words tokenize efficiently.
- Non-English languages: varies widely. Chinese and Japanese can use 2-3x more tokens per word equivalent.
What does 750,000 words cost?
Since 750,000 words equals 1 million tokens, the cost is simply each model's per-million-token price:
| Model | Read 750K words (input) | Write 750K words (output) |
|---|---|---|
| Gemini 3 Flash | $0.50 | $3 |
| GPT-5.6 Luna | $1 | $6 |
| Claude Haiku 4.5 | $1 | $5 |
| Gemini 3.5 Flash | $1.50 | $9 |
| Claude Sonnet 5 | $2 | $10 |
| Gemini 3.1 Pro | $2 | $12 |
| GPT-5.6 Terra | $2.50 | $15 |
| GPT-5.6 Sol | $5 | $30 |
| GPT-5.5 | $5 | $30 |
| Claude Opus 4.8 | $5 | $25 |
| Claude Fable 5 | $10 | $50 |
| GPT-5.5 Pro | $30 | $180 |
Prices updated 2026-07-28.
Frequently asked questions
- How many words is 1 million tokens?
- Approximately 750,000 English words, using the standard 1 token = 0.75 words conversion.
- How do you convert tokens to words?
- Multiply the number of tokens by 0.75. For words to tokens, multiply by 1.33.
- Why are tokens different from words?
- AI models break text into tokens, not words. A token can be a full word, part of a word, or punctuation. Common words like "the" are single tokens, while longer words become multiple tokens.
- Is 1 token always 0.75 words?
- No. The 0.75 ratio is an average for English prose. Code averages ~0.5 words per token. Non-English languages vary widely.