Track 1

Absolute Accuracy

Ranked purely on the highest reported test accuracy, independent of model size, compute, or memory footprint. A team's result is its single best submission. There is no penalty for the number of attempts.

Submission Accuracy

Each point is one submission, positioned by the date it was made.

Submission Results

Teams are shown once, ranked by their best accuracy. Expand a row to see a team's other submissions.

Track 2

Compute Efficiency

Ranks the trade-off between accuracy, compute (effective MACs + effective ACs), and memory footprint. Two independent scoring strategies were developed and are both reported below for transparency.

Evaluation Strategy 1

NeuroEfficiency Score (NES)

Accuracy minus a log-penalty for compute+footprint, self-normalized per submission

    Accuracy vs. Weighted-Ops

    Log-scaled x-axis. Ringed points sit on the Pareto front — no submission beats them on accuracy, ops, and footprint simultaneously.

    Best NES per team

    Each team's highest-scoring submission.

    Results — Eval Strategy 1

    Evaluation Strategy 2

    Normalized Balanced Score (NBS)

    Accuracy x Savings Score, normalized against the pool, but skewed by outliers

      Accuracy vs. Footprint

      Log-scaled x-axis.

      Best NBS per team

      Each team's highest-scoring submission.

      Results — Eval Strategy 2