Horangi 4 - Korean Hate Speech: leaderboard
Korean hate-speech classification questions, as sampled by the Horangi Korean LLM Leaderboard 4 (W&B Korea). Accuracy (%), higher is better.
Metric: Accuracy (%). Source: horangi.ai. Saturation forecast: Not forecast. 105 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Fable 5 (High) | 78 | #20 (Claude Fable 5) |
| 2 | Gemma 3 27B (IT) | 74 | #528 |
| 3 | EXAONE-3.0-7.8B-Instruct | 71 | #1146 |
| 4 | Grok 4.1 Fast (Non-reasoning) | 70 | #227 (Grok 4.1 Fast) |
| 5 | DeepSeek V3.2 (High) | 70 | #210 (DeepSeek V3.2) |
| 6 | Grok 4.20 | 69 | #179 |
| 7 | Mistral Medium 3.5 | 69 | #375 |
| 8 | Claude Opus 4.6 (High) | 69 | #67 (Claude Opus 4.6) |
| 9 | GLM-5-Turbo | 69 | #121 |
| 10 | Qwen 3.5 397B A17B | 67 | #154 |
| 11 | Mistral Small 4 | 67 | #522 |
| 12 | Claude Sonnet 4.6 (High) | 67 | #94 (Claude Sonnet 4.6) |
| 13 | A.X 4.0 | 67 | #436 |
| 14 | Gemma 3 12B (IT) | 66 | #672 |
| 15 | Qwen 3.7 Max | 66 | #85 |
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-korean-hate-speech · How It Works · Data refreshed daily, snapshot 2026-11-05.