Horangi 4 - KorQuAD 1.0: leaderboard
KorQuAD 1.0 (squad_kor_v1), Korean extractive reading comprehension, as sampled by the Horangi Korean LLM Leaderboard 4 (W&B Korea). F1 x100 of extracted answers, higher is better.
Metric: F1 (x100). Source: horangi.ai. Saturation forecast: Estimated already saturated. 105 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3.7 Max | 90.34 | #85 |
| 2 | Claude Sonnet 4.5 (High) | 90.05 | #148 (Claude Sonnet 4.5) |
| 3 | Claude Sonnet 5 (High) | 89.1 | #65 (Claude Sonnet 5) |
| 4 | Qwen 3.6 Plus (Preview) | 88.89 | |
| 5 | Claude Opus 4.1 (20250805) | 88.85 | #139 |
| 6 | Gemini 3.1 Flash Lite (Preview) | 88.51 | #201 |
| 7 | Gemini 3.1 Pro (Preview) (High) | 88.18 | #64 (Gemini 3.1 Pro (Preview)) |
| 8 | Claude Opus 4.5 (20251101) (High) | 87.68 | #81 (Claude Opus 4.5 (20251101)) |
| 9 | GPT-4.1 Mini | 87.26 | #363 |
| 10 | EXAONE 3.5 7.8B Instruct | 87.23 | #856 |
| 11 | Gemini 3 Flash (Preview) | 87.19 | #90 |
| 12 | GPT-4o (2024-11-20) | 87.17 | #392 |
| 13 | Gemma 4 31B (IT) (Thinking) | 87.07 | #178 (Gemma 4 31B (IT)) |
| 14 | Claude Opus 4.6 (High) | 87.03 | #67 (Claude Opus 4.6) |
| 15 | Gemini 3 Pro (Preview) (High) | 87.02 | #71 (Gemini 3 Pro (Preview)) |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=horangi-4-korquad-1-0 · How It Works · Data refreshed daily, snapshot 2026-11-05.