Benchmarks · reasoning
SimpleBench
200+ multiple-choice questions on spatio-temporal reasoning, social intelligence and adversarial trick questions where unspecialized humans outperform frontier models; accuracy averaged over 5 runs.
96 models · 97 results · data as of 2026-10-29 · simple-bench.com ↗
Leaderboard
Opus 5.5 max
All results
97 rows| # | Model | Effort | SimpleBench |
|---|---|---|---|
| 1 | Opus 5.5 | max | 88.4% |
| 2 | Fable 5.1 | max | 86.6% |
| 3 | GPT-6 Astra Pro | max | 86.5% |
| 4 | GPT-6 Astra | max | 83.6% |
| 5 | GPT-6.1 Sol | max | 82.9% |
| 6 | Gemini 3.8 Flash | max | 82.4% |
| 7 | Fable 5 | max | 81.9% |
| 8 | Muse Spark 1.3 | max | 81.8% |
| 9 | Opus 5 | max | 80.6% |
| 10 | Gemini 3.1 Pro | high | 79.6% |
| 11 | GPT-5.5 Pro | xhigh | 76.9% |
| 12 | Gemini 3.5 Flash | xhigh | 76.7% |
| 13 | Gemini 3 Pro | high | 76.4% |
| 14 | Sonnet 5.5 | max | 75.9% |
| 15 | Grok 4.6 | max | 75.9% |
| 16 | Muse Spark 1.2 | max | 74.5% |
| 17 | GPT-6 Sol | max | 73.1% |
| 18 | GPT-5.6 Sol Pro | xhigh | 71.7% |
| 19 | Qwen 3.7 Max | max | 70.4% |
| 20 | Grok 4.5 | max | 70.0% |
| 21 | GPT-5.5 | xhigh | 69.0% |
| 22 | Opus 4.6 | max | 67.6% |
| 23 | DeepSeek V4.1 Flash | max | 66.7% |
| 24 | GLM 5.3 | max | 66.2% |
| 25 | Opus 4.8 | max | 64.8% |
| 26 | GPT-5.6 Sol | xhigh | 64.8% |
| 27 | Qwen 3.6 Max | max | 63.0% |
| 28 | Qwen 3.8 2.4T-A95B | xhigh | 62.5% |
| 29 | Gemini 2.5 Pro | high | 62.4% |
| 30 | Opus 4.5 | high | 62.0% |