HAL CORE-Bench Hard: leaderboard
Princeton HAL leaderboard for CORE-Bench Hard: 270 computational reproducibility tasks from 90 scientific papers in CS, social science, and medicine. Tests ability to reproduce research results.
Metric: Accuracy (%). Source: hal.cs.princeton.edu. Saturation forecast: Around January 2027. 49 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.5 | 77.78 | #83 |
| 2 | Claude Sonnet 4.5 | 62.22 | #142 |
| 3 | Claude Opus 4.1 (20250805) | 51.11 | #132 |
| 4 | Claude Sonnet 4 (20250514) | 46.67 | #213 |
| 5 | Claude Sonnet 4.5 (High) | 44.44 | #142 (Claude Sonnet 4.5) |
| 6 | Claude Opus 4.1 | 42.22 | #140 |
| 7 | Claude Opus 4.5 (20251101) | 42.22 | #74 |
| 8 | Claude Opus 4.5 (High) | 42.22 | #83 (Claude Opus 4.5) |
| 9 | Gemini 3 Pro (Preview) (High) | 40 | #66 (Gemini 3 Pro (Preview)) |
| 10 | Claude 3.7 Sonnet (20250219) | 35.56 | #201 |
| 11 | O4 Mini (High) | 35.56 | #180 (O4 Mini) |
| 12 | GPT-4.1 (2025-04-14) | 33.33 | #250 |
| 13 | GPT-5 (Medium) | 26.67 | #98 (GPT-5) |
| 14 | O3 (Medium) | 24.44 | #124 (O3) |
| 15 | Gemini 2.5 Pro (Preview 03-25) | 22.22 | #123 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=hal-core-bench-hard · How It Works · Data refreshed daily, snapshot 2026-10-20.