MeasureTokens
LearnContact

See what your prompt costs on every model

Exact token counts, live pricing, and context-window fit across 41 models — free, in your browser, nothing stored.

Fewest tokens
—
type something
Token spread
—
across all models
Cheapest / month
—
at your volume
Max savings / month
—
cheapest vs priciest
Try:

Counts include each model's own chat-template overhead.

128 chars

Cost = uncached input × input price + cached input × the provider's cache-read price + assumed output × output price. Models without a cached rate get no discount.

41 models

ModelTokensCost
OpenAI
Anthropic
Google
xAI
Open weights

How counting works

Exact, in your browser

OpenAI and open-weights models are tokenized with their real tokenizers (tiktoken and Hugging Face), running entirely client-side. Your prompt never leaves the page.

Exact, via provider APIs

Anthropic, Google, and xAI don't publish their tokenizers, so counts come from each provider's official count-tokens endpoint — which tokenizes without running the model and never stores your prompt.

Prices that stay current

Pricing is refreshed automatically every week from a community-maintained dataset and reviewed before merging — last updated 2026-07-23. Model health is checked weekly too.

FAQ

Why do the same words cost different amounts on different models?

Every vendor trains its own tokenizer vocabulary, so identical text splits into different numbers of tokens — often 10–30% apart, more for code and non-English text. Multiply that by per-token prices that vary 100× between models and the same prompt can differ wildly in real cost.

How is monthly cost calculated?

Your prompt's input tokens × the model's input price, plus your assumed output length × its output price, times your requests per month. If part of your prompt is served from cache, set the cached-input percentage and the cached share is billed at the provider's cache-read rate.

Can a model get more expensive without its price changing?

Yes — when a vendor ships a new tokenizer. If the same prompt becomes ~30% more tokens (as with Anthropic's newest models versus their predecessors), your bill rises ~30% at an identical price per million. Because this table counts with each model's actual tokenizer, that hidden difference shows up here as a higher token count and real cost — compare the same prompt across model generations to see it, or read the effective price of frontier models for the full story.

Can I see exactly where tokens split?

Click any row to see the prompt colorized token by token, with token IDs — including each model's own chat-template overhead. For Claude, Gemini, and Grok the boundaries aren't public, but the counts are exact.