TriBench-Ko - Hallucination: leaderboard
Metric: Macro-F1 (0-100) on the TriBench-Ko Hallucination risk items (fabricated case citations, statutory provisions or facts): yes/no verification of statements about Korean court decisions, computed over the atomic binary judgments, zero-shot, temperature 0, at most 64 output tokens; higher is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 13 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5.4 | 93.3 |
| 2 | Qwen 3.5 9B | 88.9 |
| 3 | GPT-5.4 Mini | 88.3 |
| 4 | GPT-4o | 84.3 |
| 5 | Midm-2.0-Base-Instruct | 83.3 |
| 6 | Phi-4 | 82.6 |
| 7 | Ministral-3-8B-Instruct-2512 | 77.2 |
| 8 | Qwen 3 8B | 70.6 |
| 9 | Gemma 3 12B (IT) | 68.2 |
| 10 | EXAONE 3.5 7.8B Instruct | 64.8 |
| 11 | Llama 3.1 8B Instruct | 37.8 |
Interactive version: theaggregate.ai/benchmark?slug=tribench-ko-hallucination · How It Works · Data refreshed daily, snapshot 2026-10-07.