Cost and time
What each architecture costs on each model, in dollars and seconds per run and in dollars per correct run, as figures rather than tables.
A cell is one architecture on one model over the runs the filters leave. Means are over graded runs only; controls (B0, B1), stale and excluded runs do not count. $ per correct run is the cell's total spend divided by its correct runs, so a cheap architecture that is often wrong stops looking cheap. Click a mark to open the cell in the scorecard.
Dollars are list price: each run's recorded tokens priced against the bench's own table, scoring/pricing.json, pinned per model with its source and effective date. They are not what a contributor was billed; discounts, credits and OpenRouter routing vary per account and would make cost a property of the contributor instead of the run. A model without a row cannot enter the dataset.