HAL GAIA Level 3 — leaderboard
Metric: Accuracy (%). Source: hal.cs.princeton.edu. 32 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Sonnet 4.5 | 65.39 |
| 2 | Claude Haiku 4.5 (20251001) | 61.54 |
| 3 | Claude 3.7 Sonnet (High) | 57.69 |
| 4 | O4 Mini (Low) | 53.85 |
| 5 | Claude Opus 4.1 (High) | 53.85 |
| 6 | Claude 3.7 Sonnet (20250219) | 46.15 |
| 7 | GPT-5 (Medium) | 46.15 |
| 8 | O4 Mini (High) | 46.15 |
| 9 | Claude Sonnet 4.5 (High) | 46.15 |
| 10 | Claude Opus 4 (20250514) | 42.31 |
| 11 | Claude Opus 4.1 (20250805) | 42.31 |
| 12 | GPT-4.1 (2025-04-14) | 34.62 |
| 13 | O3 (Medium) | 23.08 |
| 14 | DeepSeek R1 | 15.38 |
| 15 | Gemini 2.0 Flash (001) | 11.54 |
Interactive version: theaggregate.ai/benchmark?slug=hal-gaia-level-3 · How the rankings work · Data refreshed daily, snapshot 2026-07-22.