KRETA: leaderboard
Vision-language models answer 2,577 Korean multiple-choice questions about text-rich images (signs, documents, menus, exam pages) across 15 domains; every model ran all items and the score is the share answered correctly. Unlike English text-VQA sets, the text, domains and the CSAT exam items are Korean.
Metric: Accuracy on all 2,577 Korean text-rich VQA items (%). Source: github.com. Saturation forecast: Estimated already saturated. 17 models tracked. Updated 17 Nov 2026.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4o (2024-11-20) | 84.6 | #864 |
| 2 | Claude 3.5 Sonnet (20241022) | 80.5 | #648 |
| 3 | A.X-4.0-VL-Light | 78 | #2575 |
| 4 | VARCO-VISION-2.0-14B | 75.4 | #1794 |
| 5 | kanana-1.5-v-3B-instruct | 75 | #3953 |
| 6 | VARCO-VISION-14B | 72.3 | #1854 |
| 7 | Qwen 2.5 VL 3B Instruct | 71.8 | #2648 |
| 8 | InternVL2.5-8B | 70.8 | #2401 |
| 9 | InternVL2.5-4B | 70.7 | #2559 |
| 10 | Qwen 2.5 VL 7B Instruct | 68.5 | #1893 |
| 11 | MiniCPM-o-2.6 | 64.3 | #2417 |
| 12 | Ovis1.6-Gemma2-9B | 58.4 | #2365 |
| 13 | DeepSeek-VL2-Small | 53.3 | #2502 |
| 14 | Ovis1.6-Llama3.2-3B | 52.2 | #3750 |
| 15 | DeepSeek-VL2-Tiny | 48.8 | #6676 |
No result here: #1 Claude Fable 5.2, #2 GPT-6.2 Astra, #3 Gemini 4 Carbon.
Interactive version: theaggregate.ai/benchmark?slug=kreta · How It Works · Data refreshed daily, snapshot 2026-11-17.