Words to Tokens & Tokens to Words - 32k, 128k, 1M
Last updated: July 2026
Planning ratios for words ↔ tokens: 200 tokens, 32k, 128k, 1 million tokens to words. Limits of fixed ratios and when to paste real text into the counter.
Words to tokens, tokens to words, and long-tails like 200 / 32000 / 128k / 1 million tokens to words are planning conversions: you have a number (word count or token budget) before you have the final string.
Why both directions exist
Writers think in words; model cards quote tokens. Words to tokens turns a brief into a token-shaped budget. Tokens to words translates 32k / 128k / 1M budgets back into approximate reading length. Both use a fixed English planning ratio (~0.75 words/token) — skewed for code and CJK.
What presets are for
Presets for 200 tokens, 32k, 128k, and 1M save typing and match common long-tail queries. They are not more accurate than a custom number — only faster for model-card scales.
Conversion is not a paste
When you have the real prompt, use the token counter. Converters answer “about”; the counter answers “this string.” Billing still follows the provider API.
Pair with the context window tool
Convert for intuition, then paste the draft into the context window calculator for fill against 128k and friends. Plan with numbers; send with a paste.
CJK caveat
Chinese without spaces breaks “word” counts. Prefer characters and tokens, or paste the final draft — do not trust word conversion alone.