Horangi 4 - Ko-HLE: leaderboard
A Korean translation of a Humanity's Last Exam subset, as run by the Horangi Korean LLM Leaderboard 4 (W&B Korea). Accuracy (%) of graded answers, higher is better.
Metric: Accuracy (%). Source: horangi.ai. Saturation forecast: Around March 2028. 105 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Fable 5 (High) | 53.06 | #20 (Claude Fable 5) |
| 2 | GPT-5.6 Sol (Max) | 52 | #22 (GPT-5.6 Sol) |
| 3 | Grok 4.5 | 50 | #54 |
| 4 | GPT-5.6 Terra (Max) | 47 | #40 (GPT-5.6 Terra) |
| 5 | GPT-5.4 (2026-03-05) (xHigh) | 47 | #63 (GPT-5.4 (2026-03-05)) |
| 6 | GPT-5.5 (xHigh) | 46.46 | #33 (GPT-5.5) |
| 7 | Gemini 3.5 Flash | 45 | #66 |
| 8 | Gemini 3.1 Pro (Preview) (High) | 45 | #64 (Gemini 3.1 Pro (Preview)) |
| 9 | Qwen 3.7 Max | 42.39 | #85 |
| 10 | Gemini 3 Flash (Preview) | 41 | #90 |
| 11 | GPT-5.6 Luna (Max) | 40 | #79 (GPT-5.6 Luna) |
| 12 | Claude Opus 4.6 (High) | 37 | #67 (Claude Opus 4.6) |
| 13 | Gemini 3 Pro (Preview) (High) | 37 | #71 (Gemini 3 Pro (Preview)) |
| 14 | GPT-5.2 (2025-12-11) (xHigh) | 36.47 | |
| 15 | Gemma 4 31B (IT) (Thinking) | 35 | #178 (Gemma 4 31B (IT)) |
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-hle · How It Works · Data refreshed daily, snapshot 2026-11-05.