Horangi 4 - Ko-MT-Bench: leaderboard
Ko-MT-Bench, a Korean MT-Bench with two-turn answers rated 1-10 by an LLM judge, as run by the Horangi Korean LLM Leaderboard 4 (W&B Korea). The board reads the judge mean from each run's detail table (the summary divides it by 100 on most runs) and reports rating x10 (10-100); higher is better.
Metric: Judge rating (1-10, x10). Source: horangi.ai. Saturation forecast: Estimated already saturated. 105 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.6 (High) | 90.13 | #67 (Claude Opus 4.6) |
| 2 | Gemini 3.5 Flash | 89.63 | #66 |
| 3 | GPT-5.1 (2025-11-13) | 89.63 | #107 |
| 4 | Claude Opus 4.5 (20251101) (High) | 89.63 | #81 (Claude Opus 4.5 (20251101)) |
| 5 | Gemini 3 Flash (Preview) | 89.62 | #90 |
| 6 | Claude Sonnet 4.6 (High) | 89.5 | #94 (Claude Sonnet 4.6) |
| 7 | Claude Sonnet 4.5 (High) | 89.38 | #148 (Claude Sonnet 4.5) |
| 8 | GPT-OSS-120B (High) | 89.25 | #346 (GPT-OSS-120B) |
| 9 | Gemini 3.1 Pro (Preview) (High) | 89.25 | #64 (Gemini 3.1 Pro (Preview)) |
| 10 | Claude Sonnet 5 (High) | 89.25 | #65 (Claude Sonnet 5) |
| 11 | Qwen 3.7 Max | 89.13 | #85 |
| 12 | Gemini 3 Pro (Preview) (High) | 89.13 | #71 (Gemini 3 Pro (Preview)) |
| 13 | Gemini 3.1 Flash Lite (Preview) | 89 | #201 |
| 14 | MiMo-V2-Pro | 89 | #177 |
| 15 | Claude Opus 4.1 (20250805) | 88.88 | #139 |
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-ko-mt-bench · How It Works · Data refreshed daily, snapshot 2026-11-05.