SimpleBench — leaderboard
Evaluates 'linguistic adversarial robustness' - ability to answer trick questions. 200+ questions where everyday human reasoning (83.7%) still outperforms frontier LLMs on spatio-temporal reasoning and social intelligence.
Metric: Score (AVG@5). Source: simple-bench.com. Status: saturation imminent. 87 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Human Expert | 95.4 |
| 2 | Median Human | 83.7 |
| 3 | Claude Fable 5 | 81.9 |
| 4 | Gemini 3.1 Pro (Preview) | 79.6 |
| 5 | GPT-5.5 Pro | 76.9 |
| 6 | Gemini 3.5 Flash | 76.7 |
| 7 | Gemini 3 Pro (Preview) | 76.4 |
| 8 | Qwen 3.7 Max | 70.4 |
| 9 | Grok 4.5 | 70 |
| 10 | GPT-5.5 | 69 |
| 11 | Claude Opus 4.6 | 67.6 |
| 12 | Claude Opus 4.8 | 64.8 |
| 13 | GPT-5.6 Sol (xHigh) | 64.8 |
| 14 | Qwen 3.6 Max Preview | 63 |
| 15 | Gemini 2.5 Pro (Preview 06-05) | 62.4 |
Interactive version: theaggregate.ai/benchmark?slug=simplebench · How the rankings work · Data refreshed daily, snapshot 2026-07-22.