HAL AssistantBench — leaderboard
Princeton HAL leaderboard for AssistantBench: web assistance tasks requiring multi-step browsing, information gathering, and synthesis.
Metric: Accuracy (%). Source: hal.cs.princeton.edu. Status: saturation imminent. 15 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | O3 (Medium) | 38.81 |
| 2 | GPT-5 (Medium) | 35.23 |
| 3 | O4 Mini (Low) | 28.05 |
| 4 | O4 Mini (High) | 23.84 |
| 5 | GPT-4.1 (2025-04-14) | 17.39 |
| 6 | Claude 3.7 Sonnet (20250219) | 16.69 |
| 7 | Claude Opus 4.1 (High) | 13.75 |
| 8 | Claude 3.7 Sonnet (High) | 13.08 |
| 9 | Claude Sonnet 4.5 (High) | 11.8 |
| 10 | DeepSeek R1 0528 | 8.75 |
| 11 | Claude Opus 4.1 (20250805) | 7.26 |
| 12 | Claude Sonnet 4.5 | 7.09 |
| 13 | Gemini 2.0 Flash (001) | 2.62 |
| 14 | DeepSeek V3 (0324) | 2.03 |
| 15 | DeepSeek R1 | 0 |
Interactive version: theaggregate.ai/benchmark?slug=hal-assistantbench · How the rankings work · Data refreshed daily, snapshot 2026-07-22.