RINoBench - Novelty Score MAE: leaderboard
Metric: Mean absolute error (score points, 0-4) between the predicted and gold 1-5 novelty scores on the RINoBench test split (one fifth of the 1,381 ICLR 2022-2023 research ideas, gold scores from human reviewers), zero-shot with the five-point novelty rubric and the idea's related works; lower is better. Source: arxiv.org. Saturation forecast: Around January 2028. 7 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 | 0.93 | #91 |
| 2 | O3 | 0.93 | #121 |
| 3 | GPT-OSS-120B | 0.96 | #330 |
| 4 | DeepSeek R1 | 0.99 | #245 |
| 5 | Llama 3.1 8B | 1 | #1139 |
| 6 | Llama 4 Scout | 1.01 | #646 |
| 7 | Llama 3.3 70B Instruct | 1.04 | #520 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=rinobench-novelty-score-mae · How It Works · Data refreshed daily, snapshot 2026-10-11.