About TokenPad
Why this exists
Anyone building on a language model API hits the same wall within a week: the bill is denominated in a unit you cannot easily see. Tokens are invisible in your editor, they vary by model, they vary by language, and the tooling that claims to count them mostly divides your character count by four and rounds.
TokenPad exists to make that unit legible. Count a prompt properly, see the price, see whether it fits, and see where the tokens went — with an explicit statement of how confident the number is.
The one rule
Never present an estimate as a measurement. 42 models are tracked here. For those whose tokenizers are published, the real encoder runs in your browser and the count is exact. For the rest, no browser-capable tokenizer exists, so the figure is derived and labelled an estimate everywhere it appears — on the readout, in every table, and in the structured data.
It would be easy to drop the labels and look more capable than we are. That is precisely the thing worth not doing: a tool that quietly reports an approximation as a fact is worse than no tool, because it produces confident wrong budgets. The full derivation is on the methodology page.
How the data is maintained
Prices for OpenAI, Anthropic, Google, DeepSeek, xAI, Mistral AI are read from each provider’s own documentation, never from an aggregator and never from memory. Each entry stores the source URL and the date it was read, both published in the model price table, and rows older than 45 days are flagged automatically. If a figure is wrong, that is a bug — report it and it gets fixed.
How it is built
Static HTML with no application server and no database. Nothing you type is transmitted, because there is nowhere for it to go. No web fonts, no tracking scripts, no cookie wall in front of the tool. The tokenizer weights load only when you actually use a tool, so reading a page costs you almost nothing.
How it is funded
The intention is display advertising, which is why the layout reserves space for ad units from the outset — retrofitting them is what makes sites jump around while they load. No advertising runs today. When it does, it will sit below the tool rather than in front of it, and no tool will ever go behind a login, a paywall or an email form.
Independence
No affiliation with OpenAI, Anthropic, Google, DeepSeek or any other provider. No sponsored placements, and no model ranked by anything other than the numbers in the table. Model names and trade marks belong to their owners and appear here for identification and comparison only.