SimpleBench
Classic NLP · 2024-08-20
SimpleBench tests everyday commonsense reasoning with more than 200 multiple-choice questions spanning spatial and temporal reasoning, social intelligence, and adversarial trick questions. The official MCQ leaderboard reports accuracy averaged across five runs (AVG@5).
Top models (higher is better)
| Model | Score |
|---|---|
| Highest Human Score (SimpleBench) | 95.4 |
| Claude Opus 5.5 | 88.4 |
| Claude Fable 5.1 | 86.6 |
| GPT-6 Astra Pro | 86.5 |
| Human Baseline (SimpleBench) | 83.7 |
| GPT-6 Astra | 83.6 |
| GPT-6.1 Sol | 82.9 |
| Gemini 3.8 Flash | 82.4 |
| Claude Fable 5 | 81.9 |
| Muse Spark 1.3 | 81.8 |
| Claude Opus 5 | 80.6 |
| Gemini 3.1 Pro Preview | 79.6 |
| GPT-5.5 Pro | 76.9 |
| Gemini 3.5 Flash | 76.7 |
| Gemini 3 Pro Preview | 76.4 |
| Grok 4.6 | 75.9 |
| Claude Sonnet 5.5 | 75.9 |
| Muse Spark 1.2 | 74.5 |
| GPT-6 Sol | 73.1 |
| GPT-5.6 Sol Pro | 71.7 |
| Qwen3.7-Max (source unspecified) | 70.4 |
| Grok 4.5 | 70.0 |
| GPT-5.5 | 69.0 |
| Opus 4.6 | 67.6 |
| DeepSeek V4.1 Flash | 66.7 |
| GLM-5.3 | 66.2 |
| Opus 4.8 | 64.8 |
| GPT-5.6 Sol | 64.8 |
| Qwen3.6-Max-Preview | 63.0 |
| Qwen3.8 2.4T A95B | 62.5 |
| Gemini 2.5 Pro | 62.4 |
| Opus 4.5 | 62.0 |
| Opus 4.7 | 61.7 |
| GPT-5 Pro | 61.6 |
| Gemini 3 Flash Preview | 61.1 |
| DeepSeek-V4-Flash-0731 | 61.1 |
| Kimi K3 | 60.7 |
| Sonnet 5 | 60.6 |
| Grok 4 | 60.5 |
| Qwen3.8 27B | 60.2 |