OSQ-Bench: leaderboard
OSQ leaderboard averaging multiple-choice accuracy across commonsense-reasoning datasets (WinoGrande, ARC, MMLU, PIQA, RACE).
Metric: Average. Source: huggingface.co. Saturation forecast: Estimated already saturated. 7 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4o | 70.15 | #343 |
| 2 | GPT-4 Preview (1106) | 65.93 | #418 |
| 3 | Claude 3 Opus | 62.53 | #571 |
| 4 | Mistral Large | 60.48 | #710 |
| 5 | GPT-3.5 | 60.32 | #818 |
| 6 | Gemini 1.0 Pro | 54.06 | #933 |
| 7 | Llama 3 70B Instruct | 52.92 | #634 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=osq-bench · How It Works · Data refreshed daily, snapshot 2026-10-20.