InsanityBench — leaderboard
10 handcrafted cryptic puzzles mixing stories, images, and cryptic number sequences that probe lateral thinking and creative leaps. Tasks require visual interpretation (e.g. recognizing objects in images) combined with wordplay and cross-domain reasoning.
Metric: Total Score. Source: robinhaselhorst.com. Status: years away from saturation. 8 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 15 |
| 2 | GPT-5.2 (High) | 10 |
| 3 | GPT-5.4 (High) | 10 |
| 4 | Claude Opus 4.6 (Medium) | 10 |
| 5 | Grok 4 | 5 |
| 6 | Gemini 3 Flash (High) | 5 |
| 7 | Grok 4.1 Fast (Reasoning) | 0 |
| 8 | GPT-5.2 (Low) | 0 |
Interactive version: theaggregate.ai/benchmark?slug=insanitybench · How the rankings work · Data refreshed daily, snapshot 2026-07-22.