Intelligence, price and speed — in four charts
The whole model landscape read in one screen: who scores highest, what it costs per million tokens, and how fast it answers. Every point is a measurement published by Artificial Analysis, refreshed here every morning. Nothing on this page is our opinion.
The composite intelligence index, top 16. Variants of the same model at different reasoning efforts are counted once.
Each dot is a model — reasoning-effort variants appear separately here, because the trade-off between them is exactly what this chart shows. Left is cheaper, higher is smarter — so the best value sits top-left. The price is Artificial Analysis’ blended rate, three input tokens for one output token. The horizontal scale is logarithmic because prices span a factor of eight hundred.
The best intelligence index available on any given day, from the first model in the set to today. The line only ever goes up: it steps whenever a model beats every predecessor. Twenty-seven steps so far.
One bar per company: their highest-scoring model, whatever it costs. This is the chart that answers « who is actually competitive » without reading a table of six hundred rows.
Output tokens per second against the intelligence index. A model can be brilliant and slow, or quick and shallow — this is where the trade-off shows.
Median output tokens per second, top 14. Speed is measured on far fewer models than intelligence, so read this as a shortlist, not a full ranking.