Intelligence bar

"Gemini 3 Pro equivalent intelligence" is defined as any commercially available model scoring the same as Gemini 3 Pro on ECI or better

Currently this is 154, but the bar will track any changes in ECI methodology that might change the 3 Pro score.

Cost

Cost is defined as the price for 1m input tokens plus 1m output tokens for the cheapest model that passes the quality bar, at the generally-available API pricing tier for that model.

At time of writing, 3-pro preview costs $2/Mtok input plus $12/Mtok output, so $14 is the reference cost basis.

E.g. a future qualifying model that costs $1/Mtok input plus $3/MTok output would resolve as 4/14 = 0.286... thus 0.2x-0.5x.

Fine print

• Eligible models must be commercially available over a hosted API which is widely accepted as 'production quality'.

• In the event that ECI no longer exists or is not kept up to date, I'll define equivalence based on the most widely accepted and standard general capability benchmark available.

• In the event of exactly hitting the threshold between two buckets, the cost will count as being in the higher of the two buckets.