Kaleidoscope - Text-only: leaderboard

Each vision-language model answers the 9,452 Kaleidoscope exam questions that have no image, in their original language across 18 languages; the score is accuracy averaged over languages with unparseable answers counted wrong. It isolates in-language subject knowledge from the image questions on the Multimodal board.

Metric: Accuracy (%). Source: arxiv.org. Saturation forecast: Estimated already saturated. 8 models tracked. Updated 17 Nov 2026.

Top models

#ModelScoreOverall rank
1Qwen 2.5 VL 72B Instruct60#849
2Qwen 2.5 VL 32B Instruct53.77#1074
3Aya Vision 32B44.73#2271
4Qwen 2.5 VL 7B Instruct43.91#1893
5aya-vision-8B39.27#4874
6Qwen 2.5 VL 3B Instruct38.51#2648
7Pangea-7B37.84#4993
8Molmo 7B-D35.12#5984

No result here: #1 Claude Fable 5.2, #2 GPT-6.2 Astra, #3 Gemini 4 Carbon.

Interactive version: theaggregate.ai/benchmark?slug=kaleidoscope-text-only · How It Works · Data refreshed daily, snapshot 2026-11-17.