TW-LegalBench - Statute Citation Accuracy: leaderboard
Statute citation in TW-LegalBench (2026), a Traditional Chinese benchmark built from Taiwan official legal examinations and court judgments: the percent of criminal judgment predictions that cite the applicable statutory provisions exactly, down to paragraph and item. Higher is better.
Metric: Statute accuracy (%): applicable provisions cited exactly (article, paragraph, subparagraph, item) in 535 criminal judgment predictions, zero-shot chain-of-thought. Source: arxiv.org. Saturation forecast: Around 2029. 12 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 4.5 | 5.7 | #250 |
| 2 | GPT-5.2 | 3.9 | #176 |
| 3 | GPT-5 | 2.6 | #167 |
| 4 | Qwen 3 235B A22B | 2.4 | #623 |
| 5 | GPT-4o (2024-08-06) | 2 | #740 |
| 6 | Llama 3.1 405B | 1.1 | #1076 |
| 7 | Qwen 2.5 7B | 0.6 | #2703 |
| 8 | GPT-OSS-20B | 0.4 | #1147 |
| 9 | GPT-OSS-120B | 0.3 | #700 |
| 10 | nemotron-3-nano-30B-a3B | 0.3 | #1875 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=tw-legalbench-statute-citation-accuracy · How It Works · Data refreshed daily, snapshot 2026-11-13.