November 21, 2026
Claude Token Counter: What 1,000 Tokens Actually Costs You
Claude Token Counter: What 1,000 Tokens Actually Costs You

1,000 tokens is roughly 750 English words; at Claude Sonnet 5.5 API rates, 1,000 input tokens cost $0.002 and 1,000 output tokens cost $0.010, or one cent. Those amounts are small per request, but long pasted documents, repeated context, premium models, or high usage can make the input side add up.
Key takeaways
- 1,000 tokens ≈ 750 English words, but the estimate varies by text and tokenizer.
- Sonnet 5.5: 1,000 input tokens cost $0.002; 1,000 output tokens cost $0.010.
- A 25,000-input-token PDF prompt with an 800-token reply costs about $0.058 on Sonnet 5.5 at standard API rates.
- Use Anthropic’s token-counting tool for a model-specific estimate; do not rely on word count for an exact number.
What counts as a token in Claude?
A token is a piece of text the model processes. It might be a whole word, part of a word, punctuation, or a character sequence.
As a rough guide, 1 token is about 4 English characters or 0.75 words, so 1,000 tokens is around 750 words. This is only an approximation. The exact count varies with the text and Claude’s tokenizer, so count the actual input with Anthropic’s token-counting documentation when you need a closer estimate.
Claude API price per million tokens
Claude API prices are listed per million tokens, or MTok. Input and output are billed at different rates. Cached reads and writes have separate rates when you use prompt caching.
| Model | Input / MTok | Output / MTok | Cached read / MTok | Cached write / MTok |
|---|---|---|---|---|
| Haiku 5.5* | $0.10 | $0.50 | $0.01 | $0.125 |
| Sonnet 5.5 | $2.00 | $10.00 | $0.10 | $2.50 |
| Opus 5.5 | $4.00 | $20.00 | $0.20 | $5.00 |
| Fable 5.2 | $8.00 | $40.00 | $0.20 | $10.00 |
*Haiku 5.5 rates shown apply to prompts of 100,000 tokens or fewer. For prompts above 100,000 tokens, its input rate is $0.50/MTok and output rate is $2.50/MTok. Prices shown are standard API rates; cached rates apply when prompt caching is used. Anthropic’s pricing page and API pricing documentation checked November 2026.

Five real prompts, counted and priced
These are realistic examples, not exact counts for specific email, code, or PDF files. Counts are approximate; your content, formatting, conversation history, and Claude’s tokenizer can change them. Costs use standard API input and output rates above, without caching or additional tools.
| Example prompt | Approximate tokens (input / output) | Haiku 5.5 | Sonnet 5.5 | Opus 5.5 | Fable 5.2 |
|---|---|---|---|---|---|
| “What’s the difference between TCP and UDP?” | 10 / 150 | $0.000076 | $0.001520 | $0.003040 | $0.006080 |
| Short email rewrite request with email pasted | 250 / 200 | $0.000125 | $0.002500 | $0.005000 | $0.010000 |
| Outline for a 1,200-word blog article | 700 / 900 | $0.000520 | $0.010400 | $0.020800 | $0.041600 |
| Code review with a 300-line file pasted | 4,000 / 600 | $0.000700 | $0.014000 | $0.028000 | $0.056000 |
| Summary of a 20-page PDF | 25,000 / 800 | $0.002900 | $0.058000 | $0.116000 | $0.232000 |
The formula is:
Cost = (input tokens ÷ 1,000,000 × input rate) + (output tokens ÷ 1,000,000 × output rate)
For the email rewrite on Sonnet 5.5: (250 ÷ 1,000,000 × $2) + (200 ÷ 1,000,000 × $10) = $0.0025.
For the PDF summary on Opus 5.5: (25,000 ÷ 1,000,000 × $4) + (800 ÷ 1,000,000 × $20) = $0.116.
Notice two things. For the one-line question, every model costs less than a cent, but the Fable estimate is about 80 times the Haiku estimate. For the PDF, the large input is the bigger part of the bill: sending 25,000 tokens costs more than generating an 800-token summary on every listed model.
If you want a live estimate while drafting, Token Wise shows estimated tokens and prompt-level cost in the Claude prompt box, as well as in ChatGPT, Perplexity, Gemini, Grok, Poe, OpenRouter, ChatHub, TypingMind, and You.com. It can rewrite a padded prompt in one click and show an original-versus-optimized diff. This is an estimate, not a view of your Anthropic account or invoice. Prompts stay on your device unless you click Optimize, and are never stored or logged. See how it works.
Why long pasted documents change the math
A long document adds input tokens before Claude starts its reply. At standard rates, the PDF example’s 25,000 input tokens cost $0.05 on Sonnet 5.5; its estimated 800-token answer adds another $0.008.
If you resend the same document or other large context, prompt caching can reduce the rate for eligible cached input. A cache write has its own rate, and caching needs to be used in your API setup; it is not an automatic discount on every Claude chat. On Sonnet 5.5, for example, cached reads are priced at $0.10/MTok versus $2/MTok for standard input.
Haiku 5.5 has a higher input rate for prompts above 100,000 tokens: $0.50/MTok rather than $0.10/MTok. For non-urgent API work, Anthropic’s Batch processing offers a 50% discount on input and output tokens. Check the Batch processing documentation for details.

Tool use can add charges beyond token rates. For example, Anthropic lists web search at $10 per 1,000 searches and code execution at $0.05 per container-hour after 50 free hours per day. See the official pricing page for current feature charges.
Claude plans versus the API: which are you paying for?
Claude subscriptions are usage-limited plans, not per-token billing. Their effective cost per prompt depends on your usage, model, and plan limits.
| Plan | Price | Usage summary |
|---|---|---|
| Free | $0 | Limited usage every 5 hours and every week |
| Pro | $20/month, or $17/month billed annually ($200 up front) | At least 5× Free’s usage per 5-hour session |
| Max | $100/month | 5× Pro usage every 5 hours and every week |
| Ultra | $200/month | 10× Pro weekly usage; no 5-hour limit for verified accounts |
Prices and plan details checked November 2026 at claude.com/pricing. Subscription usage is not a fixed number of tokens. When you reach a limit, options may include waiting for it to reset, moving to a higher tier, or using available usage credits at API rates.

FAQ
How many tokens is 1,000 words?
About 1,300 tokens in typical English text. It is an approximation; the exact count depends on the words and tokenizer.
How do I count tokens in Claude?
Use Anthropic’s token-counting documentation to count the actual input for the model. A word-count conversion is useful for a rough estimate, not an exact token count.
Is Claude Pro cheaper than the API?
They are different billing models. Pro is a monthly subscription with usage limits; the API charges by model and tokens, so which costs less depends on how much you use Claude and which models you choose.
Does Claude have a token limit?
Yes. Claude models have context-window limits, which vary by model and can be up to 1 million tokens. Your full context includes more than the text you just typed, such as conversation history and other supplied content.
To estimate an API request, count its input and output separately and apply the model’s current rates. If you use Claude in a browser, check prompt length before pasting large documents or repeating the same context.