What it measures
The lowest price at which a given level of machine intelligence can be bought, tracked continuously. Three capability tiers from the Artificial Analysis Intelligence Index (30, 40, 50) each get a cost floor: the cheapest blended 3:1 input:output price per million tokens among all commercially available models clearing that bar.
Why these components
Raw token price is the wrong unit and moves the wrong way. Frontier models have become MORE expensive per token as reasoning capability rose — median new-launch blended cost went from roughly $0.69 to $1.12 per million tokens over the past year. Cost compression is only visible when capability is held constant. Three tiers are tracked because compression arrives at different times for different capability levels: commodity intelligence cheapens first, frontier intelligence last.
How to read it
Higher readings mean a fixed level of capability costs less than it did across the trailing baseline. The underlying dollar figures are the more legible number and are published alongside: intelligence 40 fell from $10.00 to $0.175 per million tokens in eight months.
What it does not say
It says nothing about total AI spend, which has risen sharply even as unit costs collapsed — cheaper inference has induced far more of it. It is a list-price measure, so it does not capture negotiated enterprise pricing, committed-use discounts, or self-hosted economics. And the Artificial Analysis Intelligence Index is one benchmark among several; a different capability metric would move the tier boundaries.