Best value
—
highest efficiency
AI Efficiency Index
Which language model buys the most capability per dollar. Score is Intelligence³ ÷ cost per task, using Artificial Analysis intelligence and task cost — not sticker price per million tokens.
Highest intelligence
—
Artificial Analysis index
Cheapest capable
—
intelligence ≥ 40, lowest task cost
Fastest capable
—
intelligence ≥ 40, highest tok/s
Intelligence vs cost per task
Log cost on X. Bubble size and color follow efficiency. Orange is better value.
Rankings
Click a column to sort. Search and creator filter apply to all views.
| Intel # | Model | Creator | Cost / task | Intel | tok/s | Latency | Efficiency |
|---|
Why cube intelligence
- A cheap model that cannot reason is not efficient. Cubing the index (10³ = 1,000, 60³ = 216,000) keeps low-capability models off the top of the board.
- Capability is not linear. A 60 is not “20% better” than a 50. It clears problems the weaker model never finishes.
- Cost is the only linear term. Denominator is Artificial Analysis cost per intelligence-index task, which already includes how much a model thinks. That replaced $/1M token list price when the source leaderboard changed.
Source data: Artificial Analysis LLM leaderboard. This page is built by Masland Tech LLC. Models without a published task cost are omitted.