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. Saturation forecast: Around January 2027. 15 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | O3 (Medium) | 38.81 | #124 (O3) |
| 2 | GPT-5 (Medium) | 35.23 | #98 (GPT-5) |
| 3 | O4 Mini (Low) | 28.05 | #180 (O4 Mini) |
| 4 | O4 Mini (High) | 23.84 | #180 (O4 Mini) |
| 5 | GPT-4.1 (2025-04-14) | 17.39 | #250 |
| 6 | Claude 3.7 Sonnet (20250219) | 16.69 | #201 |
| 7 | Claude Sonnet 4.5 (High) | 11.8 | #142 (Claude Sonnet 4.5) |
| 8 | DeepSeek R1 0528 | 8.75 | #222 |
| 9 | Claude Opus 4.1 (20250805) | 7.26 | #132 |
| 10 | Claude Sonnet 4.5 | 7.09 | #142 |
| 11 | Gemini 2.0 Flash (001) | 2.62 | #360 |
| 12 | DeepSeek R1 | 0 | #251 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=hal-assistantbench · How It Works · Data refreshed daily, snapshot 2026-10-20.