Kaleidoscope - Multimodal: leaderboard
Each model answers the 11,459 Kaleidoscope exam questions that need an image, in their original language across 18 languages; the score is accuracy averaged over languages with unparseable answers counted wrong. The overall Kaleidoscope board also includes the text-only questions.
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 | 48.4 | #849 |
| 2 | Qwen 2.5 VL 32B Instruct | 44.9 | #1074 |
| 3 | Qwen 2.5 VL 7B Instruct | 36.85 | #1893 |
| 4 | Aya Vision 32B | 35.74 | #2271 |
| 5 | Qwen 2.5 VL 3B Instruct | 33.67 | #2648 |
| 6 | aya-vision-8B | 32.35 | #4874 |
| 7 | Molmo 7B-D | 31.43 | #5984 |
| 8 | Pangea-7B | 27.15 | #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-multimodal · How It Works · Data refreshed daily, snapshot 2026-11-17.