How the rankings are made
Short version: we do not test models ourselves and we do not have opinions. We collect public numbers once a day, combine them, and show them in a way anyone can read.
The three rankings
Most used
How many tokens (pieces of words) each model processed through OpenRouter, a service that routes requests from thousands of apps and developers to hundreds of AI models. It is real traffic, not a survey. The arrow next to a number shows the change against the week before.
Smartest
The Artificial Analysis Intelligence Index: an independent score that combines many difficult tests (reasoning, knowledge, maths, coding) into a single number. Higher is smarter. Models that have not been scored yet are listed without a score rather than guessed.
Best value
Intelligence points per dollar. We divide the intelligence score by a blended price: three parts input price and one part output price, per one million tokens, because most real workloads read more than they write. To stop weak models from winning only by being cheap, a model must score at least 60% of the current leader to enter this ranking. Free models are shown as free and left out of this ranking, because dividing by zero would put them all in first place.
The value map
The chart plots intelligence against price. A model is marked green when no other model is both at least as smart and at most as expensive. Those are the only models that are a rational pick on price and intelligence alone; every other model is beaten on both by at least one of them.
How often it updates
Once a day, automatically. The page shows when the last update happened. We also keep a small snapshot of each day so that price changes and new leaders can be tracked over time.
What the data cannot tell you
- Usage covers OpenRouter only. It does not include people chatting inside ChatGPT, Gemini, Claude or other first-party apps, or companies calling a provider directly. It is a strong signal of what developers choose, not a census of all AI use.
- Tokens are counted by each provider's own tokenizer, so a token of one model is not exactly the same size as a token of another.
- One score is not the whole story. A model can rank lower overall and still be the best at your specific task. Use the ranking to build a shortlist, then test.
- Prices are list prices on OpenRouter. Discounts, caching and batch prices can make real costs lower. Always confirm with the provider before you build on a price.
- Speed is the median measured across OpenRouter traffic and changes during the day.
Sources and credit
- Usage and pricing: OpenRouter (openrouter.ai/rankings). Rankings data by OpenRouter is licensed under CC BY 4.0.
- Intelligence scores: Artificial Analysis Intelligence Index (artificialanalysis.ai).
Best AI Chart is independent. It is not affiliated with or endorsed by OpenRouter, Artificial Analysis or any AI company. Rankings are calculated from data only and cannot be bought; anything sponsored is clearly labeled and sits outside the rankings.