Methodology
tokenTide shows which AI models people actually use, measured in tokens: how much text and code each model reads and writes through OpenRouter, a service that gives developers one API for hundreds of models.
Source
All usage numbers come from OpenRouter, and only from OpenRouter. We read them through its documented API, never by scraping its web pages, and republish them under its terms with the required attribution:
Source: OpenRouter (openrouter.ai/rankings), as of
Names, providers, prices and context windows come from OpenRouter's public model catalog. We collect a fresh snapshot up to three times a day; the time above is the one OpenRouter reports, not the time we fetched it.
Usage
- The 50 models with the most tokens over the last 7 complete days (UTC).
- Tile area is proportional to the square root of the volume. With a linear scale the first model is about a hundred times the last one, and most tiles would be too small to read. The square root keeps the order and still shows who dominates.
- ▲ ▼ compare the same 7 days with the previous 7. ● new marks models with no volume in the previous 7 days.
- FREE is the free variant of a model, counted separately from the paid one. PREVIEW is a pre-release or anonymous model.
- Everything outside the top 50 is grouped by OpenRouter as a long tail; the footer of the Usage view shows how much of the traffic the 50 models cover.
History
- Every day, the 10 models with the most tokens over the 7 days ending that day. The ranking is recalculated daily, so a line moves when a model gains or loses ground over a week, not on a single busy day.
- Our archive starts on August 23, 2026. Each snapshot also contains the previous 45 days of volumes, so History reaches back to July 14, 2026. A day appears only when all 7 days of its window are available.
- On August 27, 2026 our collection did not run. Usage volumes for that day were recovered from the next snapshot; prices for that day are lost.
What these numbers don't tell you
- One source, one slice of the market. OpenRouter only sees traffic that goes through OpenRouter. It doesn't see requests sent directly to OpenAI, Anthropic, Google or other providers, enterprise contracts, or models run locally. A model that is popular here is popular among OpenRouter users, mostly developers and the apps they build.
- Tokens are not people. A single coding agent can use more tokens in a day than thousands of chat users. Volume measures usage, not the number of users or how good a model is.
- Free models and promotions move the numbers. Free variants and anonymous previews offered at no cost can rise quickly and fall just as fast when the offer ends.
- No quality scores. tokenTide doesn't rank models by benchmarks or tests, only by how much they are used.
Privacy
- tokenTide uses no cookies, has no accounts and asks for no personal data.
- Visits are counted with Cloudflare Web Analytics, which doesn't use cookies, doesn't track you across sites and only gives us aggregated numbers.
- The site is hosted on Cloudflare Pages. Like any web host, Cloudflare processes your IP address to deliver the pages and protect the site from abuse.
- Fonts and scripts are served from this domain: your browser makes no requests to Google or other third parties.
- Your choices on the page (view, period) are saved only in your browser and are never sent to us.