KOFFVQA — leaderboard
Korean Open-ended Free-form VQA benchmark: 82 vision-language models evaluated on object recognition, document/table/chart understanding, OCR, and commonsense reasoning in Korean.
Metric: Overall Score (%). Source: huggingface.co. Status: saturated. 82 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 2.5 Pro (03-25) | 89.67 |
| 2 | Gemini 2.5 Flash (Preview 04-17) | 89.53 |
| 3 | Gemini 2.5 Pro (Preview 05-06) | 88.98 |
| 4 | O3 (2025-04-16) | 88.29 |
| 5 | O1 (2024-12-17) | 88.25 |
| 6 | O4 Mini (2025-04-16) | 87.85 |
| 7 | Gemini 2.5 Flash (Preview 05-20) | 86.76 |
| 8 | GPT-4.1 Mini | 85.82 |
| 9 | GPT-4.1 | 85.71 |
| 10 | GPT-4.5 (Preview) | 85.31 |
| 11 | GPT-4o (2024-11-20) | 81.96 |
| 12 | Claude 3.5 Sonnet (20241022) | 80.47 |
| 13 | Llama 4 Scout Instruct | 79.93 |
| 14 | Gemini 2.0 Flash (001) | 79.09 |
| 15 | Gemini 2.0 Flash (Preview) | 78.87 |
Interactive version: theaggregate.ai/benchmark?slug=koffvqa · How the rankings work · Data refreshed daily, snapshot 2026-07-22.