Kaleidoscope: leaderboard
Each model answers all 20,911 four-option exam questions collected in their original language across 18 languages and 14 subjects, 55% of them with an image; the score is accuracy averaged over languages, with unparseable answers counted wrong. Unlike multilingual sets built by translating English benchmarks, no item is translated.
Metric: Accuracy (%). Source: arxiv.org. Saturation forecast: Estimated already saturated. 8 models tracked. Updated 17 Nov 2026.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 2.5 VL 72B Instruct | 52.94 | #849 |
| 2 | Qwen 2.5 VL 32B Instruct | 48.21 | #1074 |
| 3 | Qwen 2.5 VL 7B Instruct | 39.56 | #1893 |
| 4 | Aya Vision 32B | 39.27 | #2271 |
| 5 | Qwen 2.5 VL 3B Instruct | 35.56 | #2648 |
| 6 | aya-vision-8B | 35.09 | #4874 |
| 7 | Molmo 7B-D | 32.87 | #5984 |
| 8 | Pangea-7B | 31.31 | #4993 |
No result here: #1 Claude Fable 5.2, #2 GPT-6.2 Astra, #3 Gemini 4 Carbon.
Interactive version: theaggregate.ai/benchmark?slug=kaleidoscope · How It Works · Data refreshed daily, snapshot 2026-11-17.