EduArt: leaderboard
EduArt (2026) tests multimodal LLMs on 871 human-authored art history questions from Italian secondary-school exercises and US AP Art History exams, in seven formats from multiple choice to word placement and error identification. The score is the macro-average over formats of format-specific accuracy, in percent, with answer-only prompting; higher is better.
Metric: Macro-average accuracy (%; mean over seven question formats of their format-specific scores: exact match for single-answer multiple choice, option-set F1 for multiple-answer choice and error identification, statement accuracy for true or false, element accuracy for positioning and blank accuracy for closed and open completion; 871 Italian and English questions, answer-only condition, one run; higher is better). Source: arxiv.org. Saturation forecast: Around February 2027. 12 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 82.8 | #94 |
| 2 | GPT-5.5 | 80 | #49 |
| 3 | Gemini 3.5 Flash | 68.9 | #101 |
| 4 | Claude Opus 4.6 | 67.3 | #102 |
| 5 | Claude Sonnet 4.6 | 64.8 | #161 |
| 6 | Qwen 3 VL 235B A22B | 64.5 | #479 |
| 7 | Gemini 3.1 Flash Lite (Preview) | 62.8 | #418 |
| 8 | Mistral Large 3 | 58.8 | #817 |
| 9 | GPT-5.4 Mini | 54.9 | #389 |
| 10 | Claude Haiku 4.5 | 53.1 | #568 |
| 11 | GPT-5.4 Nano | 40.1 | #635 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=eduart · How It Works · Data refreshed daily, snapshot 2026-11-13.