HAL SciCode: leaderboard
Princeton HAL cost-aware agent leaderboard for SciCode scientific programming tasks.
Metric: Accuracy (%). Source: hal.cs.princeton.edu. Saturation forecast: Around 2029. 33 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | O3 (Medium) | 9.23 | #124 (O3) |
| 2 | O4 Mini (Low) | 9.23 | #180 (O4 Mini) |
| 3 | Claude Opus 4.1 (20250805) | 7.69 | #132 |
| 4 | GPT-4.1 (2025-04-14) | 6.15 | #250 |
| 5 | O4 Mini (High) | 6.15 | #180 (O4 Mini) |
| 6 | GPT-5 (Medium) | 6.15 | #98 (GPT-5) |
| 7 | Claude Sonnet 4.5 | 3.08 | #142 |
| 8 | Claude 3.7 Sonnet (20250219) | 3.08 | #201 |
| 9 | Gemini 2.0 Flash (001) | 1.54 | #360 |
| 10 | Claude Sonnet 4.5 (High) | 1.54 | #142 (Claude Sonnet 4.5) |
| 11 | DeepSeek R1 | 0 | #251 |
| 12 | Claude Haiku 4.5 (20251001) | 0 | #261 |
| 13 | DeepSeek R1 0528 | 0 | #222 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=hal-scicode · How It Works · Data refreshed daily, snapshot 2026-10-20.