How Many Words Is 100K Tokens?
Short answer: about 75,000 English words, or 150 pages at 500 words each. But 100K tokens buys far less than that if the content is code, JSON or HTML — roughly 47,000 whitespace-delimited units for source code and only ~37,000 for raw markup. Below: the conversion by content type, what a 100K-token request costs on every tracked model, and why 100K is the most useful chunk size in long-document work.
The 0.75 rule, run backwards
One token averages about 0.75 English words for ordinary prose. Multiply by 100,000 and you get the headline number — then adjust for what you are actually sending.
What 100K tokens gives you, by content type
Content type changes the answer more than any other factor. For code, markup and structured data, "words" means whitespace-delimited units — a planning proxy, not a linguistic one.
| Content type | Words from 100K tokens | 500-word pages | Why it shrinks |
|---|---|---|---|
Plain English proseBaseline Articles, chat, support text | 71,400 – 80,000 | 143 – 160 | Common short words usually map to one token each. |
Business / technical English Reports, specs, documentation | 62,500 – 71,400 | 125 – 143 | Jargon, numbers and long words split into sub-words. |
Transcribed speech Call transcripts, dictation | 69,000 – 83,300 | 138 – 167 | Short filler words tokenize cheaply; punctuation is sparse. |
Source code Python, JavaScript, TypeScript | 41,700 – 52,600 | 83 – 105 | Indentation, operators, camelCase and symbols all split. |
JSON / structured data API payloads, configs | 37,000 – 43,500 | 74 – 87 | Keys repeat per record; each quote, brace and colon bills. |
HTML markup Scraped pages, emails | 33,300 – 40,000 | 67 – 80 | Tags and attributes add tokens that carry no words at all. |
CJK text Characters, not words | 90,900 – 105,300 chars | — | Roughly one token per character. |
Ranges invert the tokens-per-1,000-words figures published on our 1,000-words conversion table. They are planning estimates, not billing-grade: the same text meters several percent differently across OpenAI, Anthropic and Google tokenizers. Measure real documents with the Token Counter before committing to a budget.
What 100K tokens looks like in practice
Abstract token counts are hard to reason about. These are the equivalents teams actually plan against.
| Unit | 100K tokens is about… | Planning note |
|---|---|---|
English words 0.75 per token | 75,000 | A book-length manuscript in one prompt. |
500-word pages Standard manuscript page | 150 | Equivalent to a 150-page report, end to end. |
Lines of source code At ~8–12 tokens per line | 8,000 – 12,500 | A mid-size module or a small service, not a whole repo. |
Support conversations At ~2,000 tokens each | 50 | Roughly a day of tickets for a small team. |
Plain-text file size At ~4 characters per token | ≈ 400 KB | PDFs and HTML inflate well beyond this once markup is included. |
Filesize is the least reliable of these: a 400 KB PDF can meter at 150K tokens or 40K depending on whether it is text-native or scanned. Always extract and count the text.
100K in, 2K out — priced on every model
Fixed profile: 100,000 input tokens and a 2,000-token response per request, at standard per-1M rates. This is an input-heavy shape, which is why the cost pattern differs from a chat workload.
| Model | Input cost | Output cost | Per request | Per 100,000 requests | Output share of bill |
|---|---|---|---|---|---|
GPT-5.6 Sol OpenAI · flagship (promo) | $0.4000 | $0.0400 | $0.4400 | $44,000 | 9.1% |
GPT-5.6 Terra OpenAI · mid | $0.2000 | $0.0240 | $0.2240 | $22,400 | 10.7% |
GPT-5.6 LunaLowest OpenAI · budget | $0.0200 | $0.0024 | $0.0224 | $2,240 | 10.7% |
Claude Opus 5 Anthropic · flagship | $0.5000 | $0.0500 | $0.5500 | $55,000 | 9.1% |
Claude Sonnet 5 Anthropic · mid | $0.2000 | $0.0200 | $0.2200 | $22,000 | 9.1% |
Claude Haiku 4.5 Anthropic · budget | $0.1000 | $0.0100 | $0.1100 | $11,000 | 9.1% |
Gemini 3.6 Flash Google · mid (promo rate) | $0.0750 | $0.0075 | $0.0825 | $8,250 | 9.1% |
Gemini 3.5 Flash-Lite Google · budget | $0.0300 | $0.0050 | $0.0350 | $3,500 | 14.3% |
Computed at 100,000 input and 2,000 output tokens using published per-1M rates (OpenAI verified 2026-08-23, Anthropic and Google 2026-08-09 to 2026-08-16). GPT-5.6 Sol's promotional rate runs through at least Nov 21, 2026; Gemini 3.6 Flash's through Dec 31, 2026. Cheapest-to-dearest spread is 24.6x. No long-context surcharge applies at this length. Caching, batch and taxes excluded.
Convert any token budget to words and pages
100K is the default; try 1K, 10K or 1M, or type your own figure. CJK capacity is shown separately because word counts do not apply.
100K is the last cheap chunk size.
Two structural rules make 100,000 tokens a practical ceiling rather than an arbitrary round number — and both are about what you avoid, not what you gain.
It stays under OpenAI's long-context surcharge. GPT-5.6 models bill higher rates once a prompt exceeds 272,000 input tokens: input doubles and output rises 1.5x. At 100K you are at roughly 37% of that threshold. The payoff is concrete — sending 1,050,000 input tokens in one request costs $8.40 on GPT-5.6 Sol at the doubled $8.00/1M rate, while the same tokens split into 100K chunks cost $4.20. Chunking below the threshold halves the input bill.
It still fits the smallest window. Claude Haiku 4.5 has a 200,000-token context — the smallest tracked here. A 100K prompt leaves exactly 100K for system instructions, tool schemas and the response, which is workable; a 150K prompt is not. Every other tracked model sits at 1,000,000 tokens or more, where 100K is under 10% of the window.
100K tokens inside each context window
How much of each model's window a 100,000-token prompt consumes — and what is left for system prompt, tools and the answer.
| Model | Context window | 100K prompt uses | Remaining for output & scaffolding |
|---|---|---|---|
GPT-5.6 Sol / Terra / Luna OpenAI · 1,050,000 | 1,050,000 | 9.5% | 950,000 |
Claude Opus 5 / Sonnet 5 Anthropic · 1,000,000 | 1,000,000 | 10.0% | 900,000 |
Claude Haiku 4.5 Anthropic · 200,000 | 200,000 | 50.0% | 100,000 |
Gemini 3.6 Flash / Flash-Lite Google · 1,048,576 | 1,048,576 | 9.5% | 948,576 |
Context windows are shared by input and output, and each model also caps a single response separately — 128,000 tokens on GPT-5.6 and Claude Opus/Sonnet 5, 64,000 on Haiku 4.5, 65,536 on both Gemini models. Check your real document before batching with the Context Window Checker.
100K-token questions
How many words is 100,000 tokens?
About 75,000 English words, using the common 0.75 words-per-token average. For plain prose the realistic range is 71,400 to 80,000 words. Technical English gives you less — roughly 62,500 to 71,400 words — because longer words and jargon split into more sub-word tokens.
How many pages is 100K tokens?
About 150 pages at 500 words per page for ordinary prose. The page count drops sharply with structured content: roughly 94 pages of source code, 80 pages of JSON, and 73 pages of raw HTML, because markup and punctuation consume tokens that carry no words.
Does 100K tokens trigger OpenAI's long-context pricing?
No. OpenAI applies higher long-context rates only above 272,000 input tokens on GPT-5.6 models, where input doubles and output rises 1.5x. At 100,000 input tokens you are at about 37% of that threshold, so standard rates apply — which is why chunking a 1,050,000-token document into 100K pieces halves the OpenAI input bill from $8.40 to $4.20 on GPT-5.6 Sol.
Is 100K tokens a lot of context?
It is a lot of text and a small slice of a modern window. 100K tokens is a 75,000-word book, but only 9.5% of the 1,050,000-token GPT-5.6 window and only 50% of Claude Haiku 4.5's 200,000-token window — the smallest tracked here. On Haiku 4.5 a 100K-token prompt leaves just 100K tokens for system instructions, tools and the response.
How much does a 100K-token prompt cost?
For 100,000 input tokens plus a 2,000-token response, the per-request cost runs from $0.0224 on GPT-5.6 Luna to $0.5500 on Claude Opus 5 — a 24.6x spread for identical work. Because the workload is input-heavy, output is only 9–14% of the bill, the opposite of a balanced request where output dominates.
Where these numbers come from.
Conversion ratios are planning estimates derived from how modern sub-word tokenizers behave — they are not exact for any specific model, and the same text can meter several percent differently across OpenAI, Anthropic and Google. Cost figures multiply fixed token volumes by published per-1M rates taken from official provider pricing pages and re-verified weekly; they do not apply caching, batch or tiered discounts, and no long-context surcharge is modeled at 100K input because it does not apply below 272,000 tokens.
Official sources: OpenAI model docs, Anthropic pricing, Google Gemini API pricing. Always confirm the invoice before making purchasing decisions.