How the AI inference cost per million tokens is built

Methodology version 1.0. List date Oct 13, 2026.

This index

Price = each lab's published API list price, in US dollars per million tokens, for its flagship model: the most capable model it sells on its standard API. Premium tiers that charge more for the same model (faster or higher-compute modes) are excluded.
Labs: OpenAI, Anthropic and Google. Our editors name each lab's flagship and the day it took over, on the day the lab puts a new one on sale. The list is versioned: a change applies from its own date and never alters an earlier value.
Blended price = (3 × input price + output price) ÷ 4, weighting input tokens three to one, as a typical request sends more tokens than it receives.
Index = the mean of the three labs' blended prices: each lab counts once, whatever it charges.
A day is published only when all three labs' prices are known, so the index never moves because a lab dropped out.
Values before the list date are reconstructed from archived copies of the published price lists, which were taken weeks apart before 2025 and days apart since; a price change between two copies is dated to the later one. They apply the flagships named on the list date backwards and are hypothetical.

No changes since the list date.

Sources: AI labs' published API prices, and archived copies of them before the list date.