Conceptavg at 12
avg@12
An aggregate metric for reasoning evaluation in which the model samples twelve times on the same set of problems and the average per-problem accuracy is taken. It reduces variance caused by sampling randomness and reflects the model's capability distribution more stably than pass@1.
Timeline (last 90 days)
Oct 13
avg@12 means letting the model sample 12 times for the same batch of problems and taking the average of the correct rates on each problem, used to reduce the variance caused by sampling randomness.
Unverified50%