The Problem: Traditional AI leaderboards heavily track intelligence but ignore API costs, making it hard to find a good deal. Conversely, pricing aggregators rank by price but ignore intelligence. This leaves users stuck: a model with low intelligence that can't finish your task correctly is worthless, but an overpriced model is just as useless—if it is too expensive to use at scale, people simply won't afford it regardless of its score.
The Solution: This index pulls true intelligence scores from the Artificial Analysis leaderboard and correlates them against real-time blended API pricing (75% Prompt / 25% Completion) directly from OpenRouter.
The Formula: Efficiency = Intelligence³ / Blended Cost
Why Cube the Intelligence? We exponentially weight the intelligence score because capability dictates utility. If you use a linear or squared relationship, the index heavily favors ultra-cheap but "dumb" anomalies. By cubing the score, we create a massive exponential penalty for lacking capability. A model with an intelligence score of 50 is treated as 8 times more valuable than a model with a score of 25. This ensures only genuinely capable, useful models rank highly.