Claude Token Counter
Estimate how many tokens your text costs on Claude and price it across all 10 Anthropic models. Anthropic publishes no tokenizer, so this counter estimates rather than counts exactly — accurate enough to size a prompt against Claude Sonnet 5’s 1,000,000-token window, and honest about where it isn’t.
- Countingest.
- Context200K–1M
- Input$2.00
- Output$10.00
- Cached$0.20010% of input
Estimate Claude tokens
Opens on Claude Sonnet 5. Every Claude row is priced from Anthropic’s own table, so switching models re-prices your text against the real rate.
Drop files here or click to browse
Supports PDF, TXT, MD, JSON, CSV, XML — max 10MB
Count as an API request
Providers also bill the role and boundary tokens that wrap your text. Anthropic's own example — a short system prompt plus one message — costs 14 input tokens for 7 tokens of visible text.
Is this Claude token counter exact?
No, and no browser tool can be. Anthropic does not publish Claude’s tokenizer, so counts here come from a characters-per-token model calibrated on English prose. That lands within roughly 10–15% for ordinary text and drifts further on code, JSON, tables and non-Latin scripts. Treat the number as a way to size a prompt and forecast a bill, not as the figure your invoice will show.
When you need the exact count, Anthropic gives it away: the /v1/messages/count_tokens endpoint returns the token count for a complete request — system prompt, messages, tools and images included — free of charge and without consuming tokens. There is one trap worth knowing even then. Claude Opus 4.7 and later use a newer tokenizer that produces roughly 30% more tokens than 4.6 and earlier for the same text, so a count measured on an older model does not transfer to the current ones, and a prompt that fit before may not now.
Claude models and prices
10 Anthropic models, priced from Claude Haiku 4.5 at $1.00 per 1M input up to Claude Fable 5 at $10.00. Anthropic’s table is unusually regular: output costs exactly five times input on every single row, and cached reads cost exactly 10% of input on every single row. Pick a tier and you know the whole shape of its bill.
| Model | Inputper 1M tokens | Cachedper 1M tokens | Outputper 1M tokens | Context | Counting | Your cost |
|---|---|---|---|---|---|---|
| Claude Fable 5 | $10.00 | $1.00 | $50.00 | 1M | est. | — |
| Claude Opus 5 | $5.00 | $0.500 | $25.00 | 1M | est. | — |
| Claude Opus 4.8 | $5.00 | $0.500 | $25.00 | 1M | est. | — |
| Claude Opus 4.7 | $5.00 | $0.500 | $25.00 | 1M | est. | — |
| Claude Opus 4.6 | $5.00 | $0.500 | $25.00 | 1M | est. | — |
| Claude Opus 4.5 | $5.00 | $0.500 | $25.00 | 200K | est. | — |
| Claude Sonnet 5 | $2.00 | $0.200 | $10.00 | 1M | est. | — |
| Claude Sonnet 4.6 | $3.00 | $0.300 | $15.00 | 1M | est. | — |
| Claude Sonnet 4.5 | $3.00 | $0.300 | $15.00 | 200K | est. | — |
| Claude Haiku 4.5 | $1.00 | $0.100 | $5.00 | 200K | est. | — |
Rates in USD per 1M tokens, read from Anthropic’s own pricing page on August 22, 2026. Providers change prices without notice.
Claude’s context window
Claude Sonnet 5, Claude Opus 5 and Claude Fable 5 all accept 1,000,000 tokens — about 1,493 pages of prose, at roughly 670 tokens a page. Claude Haiku 4.5 holds 200K, or about 299 pages, which is still a long report or a mid-sized codebase.
A million tokens is a lot of room and a real bill. Filling Claude Sonnet 5’s window once costs $2.00 in input alone, and $5.00 on Claude Opus 5. Long conversations pay that repeatedly, because every turn resends the entire history — which is why the cached column matters more than the input column as soon as a prompt stabilises.
What Claude tokens cost
Three request shapes priced on Claude Sonnet 5, Claude Opus 5 and Claude Haiku 4.5. Because Claude bills output at exactly five times input, the third row — where the model writes a lot — moves fastest.
| Request | Claude Sonnet 5 | Claude Opus 5 | Claude Haiku 4.5 |
|---|---|---|---|
| One chat turn1,000 in / 500 out | $0.0070 | $0.018 | $0.0035 |
| Reading a long document100,000 in / 2,000 out | $0.220 | $0.550 | $0.110 |
| An agent run1,000,000 in / 50,000 out | $2.50 | $6.25 | $1.25 |
Claude Sonnet 5 is worth a second look: at $2.00 in and $10.00 out it is a third cheaper than Claude Sonnet 4.6 ($3.00 / $15.00) on both rates, and it carries the same 1,000,000-token window as Claude Opus 5. Anthropic’s cached figure is the cache-hit read rate only — writing to the cache costs more than ordinary input, so caching pays off on reuse and loses money on a prefix you send once.
Claude token counter FAQ
How many tokens is 1,000 words in Claude?
Roughly 1,330 tokens for English prose, and 1,000 tokens is about 752 words. Because Anthropic keeps its tokenizer private this is an estimate rather than a measurement, and it is least reliable exactly where you would want it most: code, structured data and languages that do not use spaces to separate words. For a firm number, call Anthropic’s /v1/messages/count_tokens endpoint.
Is this Claude token counter accurate?
It is an approximation — within about 10–15% on English prose. Anthropic publishes no tokenizer file, so there is no way to reproduce Claude’s exact segmentation in a browser, and any site claiming an exact Claude count is either using a different model’s tokenizer or guessing. The exact number is available free from the API: /v1/messages/count_tokens counts a full request, including system prompt, tools and images, without consuming tokens.
What is Claude’s context window?
1,000,000 tokens on Claude Sonnet 5, Claude Opus 5 and Claude Fable 5 — around 1,493 pages of prose. Claude Haiku 4.5 holds 200K, about 299 pages. Prompt, tool definitions, tool results and the reply all share that budget, so plan for the answer as well as the question.
Why does Claude count more tokens than other counters show?
Usually because of the tokenizer change. Claude Opus 4.7 and later use a newer tokenizer that produces roughly 30% more tokens than 4.6 and earlier for identical text, so a figure measured against an older model reads low against a current one — and a prompt that used to fit can suddenly overflow. The other common cause is overhead: system prompts, tool schemas and images all count, and none of them are in the box you pasted into.
How much does Claude cost per 1M tokens?
Claude Sonnet 5 is $2.00 in and $10.00 out; Claude Opus 5 is $5.00 and $25.00; Claude Haiku 4.5 is $1.00 and $5.00. Output is exactly five times input on every Claude model, so anything that makes the model write less — a tighter format, a lower max-tokens ceiling — saves five times what the same reduction on the prompt side would.
Does prompt caching make Claude cheaper?
Only if you reuse the prefix. Cache-hit reads cost 10% of the input rate on every Claude model — $0.200 against $2.00 on Claude Sonnet 5 — but writing to the cache costs more than ordinary input. One request that populates the cache and never returns to it is a loss; a long conversation or a document queried repeatedly is where the discount compounds.
Compare Claude against the rest
Claude’s rates sit in the middle of the market, and the spread around it is wide — the full table runs from models that cost cents per million tokens to ones that cost dollars. Compare 56 models across 11 providers, or open another brand’s counter below.
New to tokens? Read how tokenization works