SCORE Robustness (Accuracy) — leaderboard

NVIDIA SCORE evaluates robustness on AGIEval, MATH, and MMLU-Pro under benchmark perturbations, with this view averaging task accuracy.

Metric: Average Accuracy (%). Source: huggingface.co. Status: saturation imminent. 6 models tracked.

Top models

#ModelScore
1Llama 3.1 70B Instruct67.02
2Mistral Large 2 (Jul)66.97
3Llama 3 70B Instruct58.67
4Yi 1.5 34B Chat54.16
5Llama 3.1 8B Instruct48.41
6Mistral Nemo Instruct (2407)44.72

Interactive version: theaggregate.ai/benchmark?slug=score-robustness-accuracy · How the rankings work · Data refreshed daily, snapshot 2026-07-22.