benchr
AI token counter
Paste a prompt and see its token count, character count, and word count live — plus what that text costs as input across every priced model. Then hand the count to the cost calculator.
≈ 0 tokens for Claude, Gemini, Grok and other non-OpenAI models — approximate only (~4 chars/token; these families publish no browser tokenizer).
Input cost of this text
cheapest first · input price onlyLoading pricing…
Cost = tokens × each model's input price from the pricing index. Output isn't counted here — for full monthly cost use the calculator. Self-hosted models with no API price are omitted. Prices verified June 2026; see methodology.
How token counting works here
Every model splits text into tokens with its own tokenizer, so one paragraph is not the same number of tokens everywhere. benchr is honest about which numbers it can prove:
- OpenAI (exact): GPT-5, GPT-5.5, GPT-5.4 and GPT-5 Mini share the o200k_base encoding. benchr runs that exact tokenizer in your browser, so the count matches what OpenAI bills.
- Claude, Gemini, Grok, DeepSeek and others (estimate): these families don't ship a public browser tokenizer. benchr shows an approximation of ~4 characters per token and labels it clearly — never as exact.
Word and character counts are always exact. The cost figure multiplies the token count by each model's official input price; it's the cost of sending this text once as input, not a monthly bill.
Also useful
Frequently asked questions
Are the token counts exact?
For OpenAI models (GPT-5, GPT-5.5, GPT-5.4, GPT-5 Mini) the count is exact — benchr runs OpenAI's real o200k_base tokenizer in your browser. For Claude, Gemini, Grok and others there is no public browser tokenizer, so those counts are clearly-labeled estimates at roughly 4 characters per token.
Why are Claude and Gemini only estimates?
Anthropic and Google don't publish a client-side tokenizer the way OpenAI's encodings are openly available. Rather than show a fake-precise number, benchr labels those counts as approximate. Treat them as a guide, not a billing figure.
Is my text sent anywhere?
No. Counting runs entirely in your browser — the text you paste never leaves your device. benchr only fetches the public pricing file to compute cost.
How is the cost figure calculated?
It multiplies the token count by each model's published input price from the benchr model index (per million tokens). It is the input cost of the pasted text; for full monthly cost including output, use the cost calculator.