About
What this site is, how the numbers are produced, and what it deliberately does not do.
Why it exists
Token pricing is scattered across a dozen provider pages, it is quoted in different units, and it changes often. Answering a simple question like "would this job be cheaper on a smaller model" usually means opening several tabs and building a spreadsheet, and that spreadsheet is out of date within weeks.
This site does that arithmetic in one place, across every model in the catalogue at once, against prices that are refreshed daily. The eight tools exist because the same question gets asked in different shapes: per request, per month, against a subscription, against a free tier, against your own hardware, or against a fixed budget.
What "free and privacy-first" actually means
Those words get used loosely, so here they are as specifics:
- No accounts. There is no signup, no login and no email capture anywhere on the site.
- Privacy-first analytics. One tool, PostHog, counts anonymous page views and a few product events (like which model you pick) so I can see what is useful. It is cookieless, honors Do Not Track, records nothing you type, and does no session recording, heatmaps or ad tracking. Details on the privacy page.
- No cookies. The site sets none at all, and the analytics is cookieless too.
- One third party. Fonts are self-hosted, the tokenizer is bundled locally, and the price data is served from this site's own domain. The only outside request is the anonymous PostHog analytics above.
- No adverts and no affiliate links. Nothing here is ranked or promoted because someone paid.
- Free with no tier. There is no paid plan, no usage cap and no feature held back.
The full detail, including everything kept in your browser's local storage, is on the privacy page.
Where the prices come from
Rates are taken from the OpenRouter model catalogue and normalised to dollars per million tokens. A scheduled job fetches that catalogue once every 24 hours, writes a snapshot file, and that snapshot is what the site serves.
This matters for a practical reason: your browser never calls OpenRouter. However many people use the site, upstream sees exactly one request per day. It also means the tools keep working when the upstream API is down, and figures are at most about a day old.
What it does not do
- These are estimates, not invoices. Provider rates change, billing has rounding and minimums, and real usage varies. Treat every figure as a close guide.
- Text token counts are approximate. The counter uses the o200k tokenizer, which is exact for OpenAI GPT-4o era models and close for others, since every provider tokenizes differently.
- Coverage is whatever OpenRouter lists. A model not routed through OpenRouter, or released in the last day or two, may be missing.
- The provider column is the model author. It does not yet break pricing down by the host serving that model.
- Subscription plan prices are entered by hand and carry the date they were checked, so they drift faster than the API rates do.
How it is built
Static HTML, vanilla JavaScript modules and one stylesheet. No framework, no build step and no backend. The whole site is files on a CDN, which is why it loads quickly and why there is very little that could quietly start collecting data later.
Who made it
Built by Rohit Waghire, a software developer in Mumbai who builds and ships small tools solo. If something here is wrong, out of date, or missing a model you care about, the most useful thing you can do is say so on GitHub or X.