One place to see what AI is for, which models do it best, and what they costOne place to see what AI is for, which models do it best, and what they cost

The problem it solves

Model announcements arrive every week, each with its own benchmarks and its own claims. Leaderboards rank models on one signal at a time, and most of what’s written about enterprise AI is either vendor marketing or a survey of intentions. None of it answers the three questions people actually ask: what is AI being used for, which model should I trust with it, and what will it cost me in the language I work in.

This site answers all three, in one place:

• 100 leading models →←Across 6 categories, each with its strengths, weaknesses, country of origin, and release date.
• Scores you can check →←Built from independent, public sources by published rules. Every mark links to its source.
• 1000 use cases →←What organisations use AI for, by business domain, with measured adoption where it exists.
• Where AI is used →←A world map of adoption by country, and the agents enterprise platforms ship.
• What a token really costs →←The same sentence run through every tokenizer its owner publishes, in English, Arabic and Chinese. A model is priced per token, but a token is not a fixed amount of meaning — the same Arabic work can cost three times as much on one model as another at the same advertised price.

Principles

• No company pays to be listed, scored, or placed, and none sees a score before it’s published.
• Every score comes from independent public data, by rules published on the methodology page.
• Nothing is estimated. A model without independent measurement is shown unscored rather than given a guess, and a use case with no matching topic is printed with no number at all.
• Where a figure rests on converting between two different measurements, the page says so and carries the uncertainty that conversion costs, rather than hiding it in a rounder number.
• The data refreshes itself every day, and publishes only when the sources actually moved — so the date on the home page means the numbers changed, not that a job ran.

Where it’s going

The site is in beta, and the directory is deliberately a starting point rather than a finished list. The plan, as AI keeps growing:

1. 01100 models in every categoryThe directory tracks 100 models across all 6 categories together. The next step is 100 in each — language, coding, image, video, audio and search — so no category is thinner than the rest.
2. 02More languages than threeToken costs are measured in English, Arabic and Chinese. The UAE works in Hindi, Urdu, Malayalam and Tagalog as well, and each is likely penalised the same way and by a different amount. The limit is not the tokenizers but the text: it has to be genuinely parallel across every language and published by one body, which is the test the present corpus had to pass.
3. 03Fewer blanksA model is scored only when independent measurement covers 80% of its rubric’s weight, so whole categories are listed unscored rather than given a number that was guessed. Every new public source closes some of that gap.
4. 04New categories as the field opens upCategories will be added as new kinds of AI become established, rather than being fixed at 6.