HAL SciCode: leaderboard
Metric: Accuracy (%). Source: hal.cs.princeton.edu. 33 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | O4 Mini (Low) | 9.23 |
| 2 | O3 (Medium) | 9.23 |
| 3 | Claude Opus 4.1 (20250805) | 7.69 |
| 4 | Claude Opus 4.1 (High) | 6.92 |
| 5 | GPT-4.1 (2025-04-14) | 6.15 |
| 6 | GPT-5 (Medium) | 6.15 |
| 7 | O4 Mini (High) | 6.15 |
| 8 | Claude 3.7 Sonnet (High) | 4.62 |
| 9 | Claude Sonnet 4.5 | 3.08 |
| 10 | DeepSeek V3 (0324) | 3.08 |
| 11 | Claude 3.7 Sonnet (20250219) | 3.08 |
| 12 | Gemini 2.0 Flash (001) | 1.54 |
| 13 | Claude Sonnet 4.5 (High) | 1.54 |
| 14 | DeepSeek R1 | 0 |
| 15 | Claude Haiku 4.5 (20251001) | 0 |
Interactive version: theaggregate.ai/benchmark?slug=hal-scicode · How It Works · Data refreshed daily, snapshot 2026-09-05.