Free tool. No signup. Runs in your browser.
AI Token Counter and Cost Calculator
Count the tokens in any text for GPT, Claude and Gemini models and see what a request, or a month of requests, costs.
Paste text or drop a file. Counts update as you type. Everything stays in your browser.
GPT-5 / GPT-6
0
tokens, exact
GPT-4 / 3.5
0
tokens, exact
Claude
0
tokens, estimate
Gemini
0
tokens, estimate
Cost calculator
List prices as of 24 September 2026, standard tier.
Type a number to override
Cache read $0.2 per 1M
Per request
$0.00500
Per month
$5.00
Input $0.00 + output $0.00500 per request with Claude Sonnet 5.
Prices are the vendors' published list prices on 24 September 2026 and change often; check the vendor page before budgeting. Claude and Gemini counts are estimates because neither vendor publishes a browser tokenizer.
How to use the Token Counter
- 1
Paste any text
A prompt, a document, a transcript. Counts for four model families update as you type; the first count loads the tokenizer tables, later ones are instant. - 2
Pick a model
Choose the model you plan to call. The input count switches to that model's tokenizer and the list price loads into the calculator. - 3
Set the volume
Enter expected output tokens, the cached share of your input if you reuse a system prompt, and requests per month. Costs per request and per month update live.
Understand what you are paying for
Short lessons on picking the right model for each task, writing prompts that use fewer tokens, and automating without runaway costs.
A short quiz first. No payment details.
How to read a token count
The four boxes answer the same question for different model families. The o200k count applies to current OpenAI models and is exact. The cl100k count applies to older GPT-4 and GPT-3.5 models. Claude's tokenizer splits English a little finer than o200k, so the estimate is 15% higher; Gemini lands close to o200k, so the estimate is 5% higher. For budgeting purposes the estimates are within the margin that output length varies by anyway.
Why the same request costs 100 times more on one model than another
List prices span two orders of magnitude, from a few cents per million tokens for the smallest models to tens of dollars for flagship ones. Output tokens cost four to five times more than input tokens on almost every model, so a request that returns a long answer is dominated by output cost. Cached input, where a vendor offers it, cuts the price of a repeated system prompt by 75 to 90%. The calculator separates the three so you can see which lever matters for your workload.
Token counter vs the 4-characters rule
The rule of thumb that one token is four characters works for English prose and fails for everything else: code, numbers, non-Latin scripts and heavily formatted text can double the count. The tool shows the rule-of-thumb number next to the real one so you can see the gap for your own text.
More free AI tools
Prompt Optimizer
Paste a prompt, get a score out of 100 and a rewritten version that ChatGPT, Claude and Gemini answer better.
Open tool →Formula Generator
Describe what you need in plain words, fill in your ranges, and copy a working formula for Excel or Google Sheets.
Open tool →Hook Generator
Turn one topic into 20 opening lines built on proven hook patterns for LinkedIn, X, Instagram, YouTube and email.
Open tool →