TW-LegalBench - Essay Questions: leaderboard
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 April 2027. 13 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5 | 166 |
| 2 | Claude Sonnet 4.5 | 160 |
| 3 | GPT-5.2 | 160 |
| 4 | Qwen 3 235B A22B | 115 |
| 5 | GPT-OSS-120B | 53 |
| 6 | GPT-4o (2024-08-06) | 52 |
| 7 | nemotron-3-nano-30B-a3B | 39 |
| 8 | Llama 3.1 405B | 21 |
| 9 | GPT-OSS-20B | 17 |
| 10 | Qwen 2.5 7B | 17 |
Interactive version: theaggregate.ai/benchmark?slug=tw-legalbench-essay-questions · How It Works · Data refreshed daily, snapshot 2026-09-29.