AI for Education Visual Reasoning - odd one out: leaderboard
The "odd one out" subtask of AI for Education Visual Reasoning. A task from the AI for Education benchmark measuring visual reasoning over educational diagrams, charts and figures.
Metric: Accuracy (%). Source: www.fab-ai.org. Saturation forecast: Around January 2027. 67 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.6 Sol | 88.3 | #19 |
| 2 | Gemini 3.7 Flash | 84.4 | #23 |
| 3 | Gemini 3.5 Flash | 80.5 | #60 |
| 4 | Kimi K3 | 79.2 | #30 |
| 5 | Gemini 3.1 Pro (Preview) | 77.9 | #58 |
| 6 | Gemini 3 Pro | 77.9 | #80 |
| 7 | Gemini 3.6 Flash | 77.9 | #51 |
| 8 | GPT-5.5 | 76.6 | #31 |
| 9 | Qwen 3.6 Plus | 76.6 | #120 |
| 10 | Qwen 3.6 27B | 75.3 | #170 |
| 11 | Qwen 3.5 397B A17B | 74 | #146 |
| 12 | GPT-5.4 (Medium) | 74 | #79 (GPT-5.4) |
| 13 | Qwen 3.5 Plus | 72.7 | #125 |
| 14 | Gemini 3 Flash | 71.4 | #99 |
| 15 | Qwen 3.5 Flash | 64.9 | #232 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=ai-for-education-visual-reasoning-odd-one-out · How It Works · Data refreshed daily, snapshot 2026-10-20.