TW-LegalBench - Essay Questions: leaderboard
Essay questions of TW-LegalBench (2026), a Traditional Chinese benchmark built from Taiwan official legal examinations and court judgments: the number of official rubric points, out of 580 across 117 bar and judicial examination essay questions, that GPT-5 and Claude Sonnet 4.5 judges label fully correct. Higher is better.
Metric: Rubric points judged fully correct (out of 580) across the 117 bar and judicial examination essay questions, zero-shot chain-of-thought; each official rubric point is labelled correct, partial, wrong or missed by gpt-5 and claude-sonnet-4.5 judges. Source: arxiv.org. Saturation forecast: Around February 2027. 13 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 | 166 | #167 |
| 2 | Claude Sonnet 4.5 | 160 | #250 |
| 3 | GPT-5.2 | 160 | #176 |
| 4 | Qwen 3 235B A22B | 115 | #623 |
| 5 | GPT-OSS-120B | 53 | #700 |
| 6 | GPT-4o (2024-08-06) | 52 | #740 |
| 7 | nemotron-3-nano-30B-a3B | 39 | #1875 |
| 8 | Llama 3.1 405B | 21 | #1076 |
| 9 | GPT-OSS-20B | 17 | #1147 |
| 10 | Qwen 2.5 7B | 17 | #2703 |
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-essay-questions · How It Works · Data refreshed daily, snapshot 2026-11-13.