AI Octane Index · BEP Research

What a finished task actually costs.

Model prices are quoted per token, but you pay per job that comes back right — and a model that needs three tries costs three times. The same basket of work, run through the whole board, spans 273× in what it costs to get finished.

Tap or hover any dot to see which model it is.

Each dot is one model. Left is less accurate, up is more expensive per completed task, so the best place to be is bottom right. Anything inside the shaded region is beaten on both counts at once by something else on this board.

Cheapest model that can do the work, by industry

In 3 of 5 industries the cheapest deployable model is not the board's overall pick. Each industry is a slice of the same basket, so these rest on 15–27 attempts against the 96 behind the board above — a steer, not a ranking.

What one of the 32 tasks looks like

Calculate the interest accrued.

LOAN TERMS:
  - Principal: $1,000,000
  - Rate: 6.00% per annum
  - Day-count convention: Actual/360
  - Accrual period: 45 days

Respond with the interest in dollars, rounded to two decimal places, as a number
only — no commas, no currency symbol.

Every model measured

Every prompt states an output format — “as a number only” — and the grader compares exactly, so a right answer wrapped in an explanation scores zero. That is a real failure if code reads the output, and no failure at all if a person does. 7 of 33 models are affected; the worst, Llama 4 Maverick, would score 81/96 instead of 56/96 — 25 of its answers were right and rejected on presentation. The ranking above uses the strict score; decide which column is yours.