PaperBench: leaderboard
End-to-end replication of AI papers, graded against hierarchical rubrics.
Metric: Replication Score (%). Source: github.com. Status: saturated. 6 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | O1 (High) | 26 |
| 2 | Claude 3.5 Sonnet | 21 |
| 3 | O3 Mini (High) | 8.5 |
| 4 | DeepSeek R1 | 6 |
| 5 | GPT-4o | 4.1 |
| 6 | Gemini 2.0 Flash | 3.2 |
Interactive version: theaggregate.ai/benchmark?slug=paperbench · How It Works · Data refreshed daily, snapshot 2026-09-05.