Horangi 4 - ALT - Hallucination Prevention: leaderboard
Hallucination prevention category of the Horangi Korean LLM Leaderboard 4 (W&B Korea): the mean of Ko-TruthfulQA accuracy, Ko-HalluLens WikiQA correct-answer rate and Ko-HalluLens nonexistent-entity refusal rate. 0-100, higher is better.
Metric: Score (%). Source: horangi.ai. Saturation forecast: Around August 2027. 105 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) (High) | 87 | #64 (Gemini 3.1 Pro (Preview)) |
| 2 | Gemini 3.5 Flash | 82.33 | #66 |
| 3 | Qwen 3.7 Max | 78.65 | #85 |
| 4 | Grok 4.5 | 76.67 | #54 |
| 5 | Gemini 3 Pro (Preview) (High) | 76.67 | #71 (Gemini 3 Pro (Preview)) |
| 6 | GLM-5-Turbo | 75 | #121 |
| 7 | GLM-5.2 (High) | 74.46 | #80 (GLM-5.2) |
| 8 | GPT-5 Mini (2025-08-07) | 72.33 | #175 |
| 9 | GLM-5 | 71.47 | #149 |
| 10 | Claude Opus 4.8 (High) | 71.33 | #39 (Claude Opus 4.8) |
| 11 | Gemini 3 Flash (Preview) | 71 | #90 |
| 12 | Kimi K2.5 | 70.67 | #152 |
| 13 | GPT-5.1 (2025-11-13) | 70.33 | #107 |
| 14 | Qwen 3.5 397B A17B | 70.2 | #154 |
| 15 | Claude Sonnet 5 (High) | 69.67 | #65 (Claude Sonnet 5) |
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-alt-hallucination-prevention · How It Works · Data refreshed daily, snapshot 2026-11-05.