SCORE Robustness (Consistency): leaderboard
NVIDIA SCORE consistency rate measures whether models preserve answers across perturbed AGIEval, MATH, and MMLU-Pro examples.
Metric: Average Consistency Rate (%). Source: huggingface.co. Saturation forecast: Estimated already saturated. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Llama 3.1 70B Instruct | 72.39 | #561 |
| 2 | Mistral Large 2 (Jul) | 71.93 | #484 |
| 3 | Llama 3 70B Instruct | 69.11 | #634 |
| 4 | Yi 1.5 34B Chat | 58.43 | #800 |
| 5 | Llama 3.1 8B Instruct | 52.74 | #1035 |
| 6 | Mistral Nemo Instruct (2407) | 49.46 | #936 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=score-robustness-consistency · How It Works · Data refreshed daily, snapshot 2026-10-20.