TriBench-Ko - Demographic Bias: leaderboard
Metric: Macro-F1 (0-100) on the TriBench-Ko Demographic Bias risk items (paired document variants that change only a legally irrelevant protected attribute): 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: Around January 2027. 13 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5.4 | 84.5 |
| 2 | GPT-5.4 Mini | 83.9 |
| 3 | GPT-4o | 78.4 |
| 4 | Phi-4 | 78.1 |
| 5 | Qwen 3.5 9B | 75.6 |
| 6 | Qwen 3 8B | 74.2 |
| 7 | Gemma 3 12B (IT) | 73.5 |
| 8 | Ministral-3-8B-Instruct-2512 | 69.9 |
| 9 | EXAONE 3.5 7.8B Instruct | 69.9 |
| 10 | Midm-2.0-Base-Instruct | 67 |
| 11 | Llama 3.1 8B Instruct | 43.2 |
Interactive version: theaggregate.ai/benchmark?slug=tribench-ko-demographic-bias · How It Works · Data refreshed daily, snapshot 2026-10-07.