TriBench-Ko - Omission: leaderboard

The omission risk dimension of TriBench-Ko (2026), a Korean benchmark of attorney-reviewed yes/no items built from court decisions, where false statements carry outputs that leave out legally constitutive facts, precedents or provisions. The score is macro-F1 on a 0-100 scale over the atomic binary judgments, zero-shot at temperature 0; higher is better.

Metric: Macro-F1 (0-100) on the TriBench-Ko Omission risk items (outputs that leave out legally constitutive facts, precedents or provisions): 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 2029. 13 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Midm-2.0-Base-Instruct61.4#1988
2GPT-5.453.3#135
3GPT-5.4 Mini50.8#406
4Ministral-3-8B-Instruct-251250#1907
5GPT-4o48.7#751
6Qwen 3.5 9B47#828
7Phi-445.9#1931
8Qwen 3 8B42.5#1873
9Gemma 3 12B (IT)42.2#1887
10kanana-1.5-8B-instruct-250537.4#3315
11EXAONE 3.5 7.8B Instruct37.2#3410
12Llama 3.1 8B Instruct35.5#4600
13A.X-3.1-Light33.1

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=tribench-ko-omission · How It Works · Data refreshed daily, snapshot 2026-11-21.