SciDocBench - Paper-Code: leaderboard
SciDocBench workflow-centred scientific document understanding: the paper-to-code capability group over 496 evaluation instances from 124 questions across arXiv, bioRxiv and OpenReview papers. Higher is better. Source: arxiv.org/abs/2609.05141.
Metric: Score (%). Source: arxiv.org. Saturation forecast: Not forecast. 14 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Kimi K2.5 | 70.25 | #152 |
| 2 | GPT-5.6 Terra | 61.88 | #40 |
| 3 | GPT-5.6 Luna | 54.25 | #79 |
| 4 | Claude Sonnet 4.6 | 52.17 | #94 |
| 5 | Claude Opus 5 | 51.75 | #16 |
| 6 | Claude Haiku 4.5 | 51.58 | #287 |
| 7 | GPT-5.6 Sol | 46.69 | #22 |
| 8 | Claude Opus 4.8 | 44 | #39 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=scidocbench-paper-code · How It Works · Data refreshed daily, snapshot 2026-11-05.