NuclearQAv2 - Verbal: leaderboard
Short-answer concept questions of NuclearQAv2 (2026), a nuclear engineering question-answering benchmark: accuracy in percent averaged over three runs, an LLM judge deciding semantic equivalence with the reference. Higher is better.
Metric: Accuracy (%) on 283 short-answer concept questions, judged semantically equivalent to the reference by an LLM judge; mean of three runs; answers requested without reasoning text. Source: arxiv.org. Saturation forecast: Around February 2027. 9 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.2 | 82.33 | #176 |
| 2 | GPT-5.4 | 81.04 | #123 |
| 3 | NVIDIA-Nemotron-3-Super-120B-A12B-FP8 | 79.27 | #469 |
| 4 | GPT-OSS-120B | 78.33 | #700 |
| 5 | GPT-4o | 74.79 | #741 |
| 6 | Nova Pro | 70.79 | #1555 |
| 7 | Llama 3 70B Instruct | 67.49 | #1676 |
| 8 | Mistral 7B Instruct | 51.59 | #7468 |
| 9 | Llama 3 8B Instruct | 48.65 | #4612 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=nuclearqav2-verbal · How It Works · Data refreshed daily, snapshot 2026-11-13.