CogSense-Bench - Crystallized Intelligence: leaderboard
CogSense-Bench (Li et al., 2026) crystallized intelligence: 368 Bongard-RWR+ and Bongard-HOI problems that require abstracting a learned concept from real-world images. Accuracy (%); higher is better.
Metric: Accuracy (%). Source: arxiv.org. Saturation forecast: Around February 2027. 11 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | O3 | 51.4 | #224 |
| 2 | Grok 4 Fast | 45.4 | #438 |
| 3 | Gemini 2.5 Flash | 40.2 | #452 |
| 4 | GPT-5.2 | 35.9 | #172 |
| 5 | Claude Sonnet 4 | 31.3 | #397 |
| 6 | Llama 4 Scout | 29.9 | #1787 |
| 7 | Gemma 3 27B | 29.4 | #1638 |
| 8 | Intern-S1 | 21 | #616 |
| 9 | Ministral 3 14B | 10.6 | #1852 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=cogsense-bench-crystallized-intelligence · How It Works · Data refreshed daily, snapshot 2026-11-12.