The single chart that matters when choosing a model: how much measured intelligence you get for each dollar. We plot blended token price against real LMArena Elo and trace the efficient frontier — the models nothing else beats on quality at a lower price.
Each dot is a model with a real LMArena human-preference score. Up = smarter; left = cheaper. The dashed line is the efficient frontier (best quality at each price point).
12 models with a real LMArena human-preference Elo. Models without an Elo score are not plotted.
These models sit on the efficient frontier: for each, no other tracked model offers higher LMArena quality at a lower blended price.
Quality is real LMArena Elo — human-preference rankings from blind head-to-head votes, not our own estimate. Price is the blended average of input and output $/1M tokens from each model's tracked pricing.
A model is on the efficient frontier when no other model has both a higher Elo and a lower-or-equal price. The frontier is recomputed as prices and rankings update.
We deliberately do not publish an 'Elo per dollar' ratio: Elo is an interval scale, so dividing it by price is not statistically meaningful. The frontier is the honest way to compare quality against cost.