NuclearQAv2 - Numeric: leaderboard
Numeric calculation questions of NuclearQAv2 (2026), a nuclear engineering question-answering benchmark: accuracy in percent averaged over three runs, an answer counting as correct within a 15 percent relative tolerance. Higher is better.
Metric: Accuracy (%) on 206 numeric calculation questions, correct within a 15 percent relative tolerance; mean of three runs; answers requested without reasoning text. Source: arxiv.org. Saturation forecast: Estimated already saturated. 9 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-OSS-120B | 79.13 | #700 |
| 2 | NVIDIA-Nemotron-3-Super-120B-A12B-FP8 | 74.11 | #469 |
| 3 | GPT-5.4 | 72.33 | #123 |
| 4 | GPT-4o | 66.99 | #741 |
| 5 | GPT-5.2 | 66.51 | #176 |
| 6 | Nova Pro | 58.41 | #1555 |
| 7 | Llama 3 70B Instruct | 57.12 | #1676 |
| 8 | Llama 3 8B Instruct | 47.09 | #4612 |
| 9 | Mistral 7B Instruct | 3.07 | #7468 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=nuclearqav2-numeric · How It Works · Data refreshed daily, snapshot 2026-11-13.