TW-LegalBench - Constitutional Law: leaderboard
Constitutional-law questions of TW-LegalBench (2026), a Traditional Chinese benchmark built from Taiwan official legal examinations and court judgments: accuracy in percent on four-option single-answer examination questions, zero-shot with chain of thought. Higher is better.
Metric: Constitutional-law questions: accuracy (%) on four-option single-answer questions from Taiwan's 2020-2024 official legal examinations (civil service, judicial, police and professional), zero-shot chain-of-thought, JSON answer (format failures count as wrong), temperature 0 (1 for gpt-5 and gpt-5.2). Source: arxiv.org. Saturation forecast: Around February 2027. 13 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 4.5 | 89.4 | #250 |
| 2 | GPT-5 | 87.6 | #167 |
| 3 | GPT-5.2 | 84.8 | #176 |
| 4 | GPT-4o (2024-08-06) | 79.9 | #740 |
| 5 | Qwen 3 235B A22B | 78.5 | #623 |
| 6 | Llama 3.1 405B | 72.2 | #1076 |
| 7 | GPT-OSS-120B | 67.2 | #700 |
| 8 | nemotron-3-nano-30B-a3B | 63.5 | #1875 |
| 9 | Qwen 2.5 7B | 59.3 | #2703 |
| 10 | GPT-OSS-20B | 59.1 | #1147 |
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-constitutional-law · How It Works · Data refreshed daily, snapshot 2026-11-13.