CogSense-Bench - Mental Simulation: leaderboard
CogSense-Bench (Li et al., 2026) mental simulation: 150 KiVA, STARE, ARC-AGI and ARC-AGI-2 questions that require inferring and simulating visual transformations. Accuracy (%); higher is better.
Metric: Accuracy (%). Source: arxiv.org. Saturation forecast: Around 2030. 11 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.2 | 60 | #172 |
| 2 | Claude Sonnet 4 | 58 | #397 |
| 3 | Gemma 3 27B | 55.3 | #1638 |
| 4 | Llama 4 Scout | 48.7 | #1787 |
| 5 | Gemini 2.5 Flash | 40.2 | #452 |
| 6 | Ministral 3 14B | 39.3 | #1852 |
| 7 | O3 | 38.7 | #224 |
| 8 | Grok 4 Fast | 21.3 | #438 |
| 9 | Intern-S1 | 17.5 | #616 |
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-mental-simulation · How It Works · Data refreshed daily, snapshot 2026-11-12.