What does this text cost in each model?
The same work costs wildly different amounts depending on who runs it. Paste the text and see the price side by side, from cheapest to most expensive.
- Module
- Router
- Input
- Some text
- Runs on
- Your browser
- Cost
- Free, no signup
Estimated by script: Latin ≈ 4 characters per token; Chinese, Japanese and Korean cost considerably more. Public list prices, in US dollars.
Same text, same work, prices that differ by 100×
AI models charge per token, pieces of words, for input and output. What almost nobody looks at is how much this price varies: the same text that costs cents on a lightweight model costs dozens of times more on a top-tier one. For a simple task, that difference is money wasted every month.
The tool estimates the tokens in your text, adds the tokens you expect in the response and applies each model's public price. The list is sorted from cheapest to most expensive, and open-weight models are marked, because you can also run those yourself.
- 01
You paste the text
Prompt, document, transcript, anything. Nothing leaves the browser: the count happens right here.
- 02
You set the size of the response
Output usually costs three to five times more than input, so ignoring it underestimates the bill by half.
- 03
The comparison comes out sorted by price
Cheapest to most expensive, with the bar showing the real distance between them.
- 01
What is a token, exactly?
- It is the unit the model reads and writes in, usually a piece of a word. In Portuguese, a common word usually comes to between one and two tokens. In Chinese or Japanese, a single character often counts as one full token, which is why the same content costs more in those languages.
- 02
Is the count exact?
- No, and no tool that does not use each model's official tokenizer will be. Each family tokenizes differently. The estimate here is calibrated by writing system and is useful for sizing cost and context, to settle the bill, use the number the provider returns in the API response.
- 03
Is my text sent to a server?
- No. The counting and the maths happen entirely in your browser. That is why you can paste a document here that you would not paste into a tool that uploads the content.
- 04
Why do input and output cost different amounts?
- Because generating costs more than reading. The model processes the input all at once, but produces output token by token, and that generation uses more compute. That is why output generally costs three to five times more than input.
- 05
Is a more expensive model always better?
- Not for every task. Classifying, extracting a field and summarising short text are handled well by light models. Reserving the expensive model for what genuinely needs reasoning is the difference between a healthy bill and an absurd one at month’s end.
- 06
How do you choose in practice, without betting on one?
- By routing by task: the simple request goes to the cheap model, the difficult one moves up. Doing this manually on every call is impractical, it is exactly what the Router module does, deciding the destination and learning from the result.
You have seen the price gap. Who picks the destination?
Choosing the right model on every call is what the Router module does: one key for the whole market, with the decision made per task rather than per contract.