XpertBench - Law: leaderboard

The law tasks of XpertBench (2026), open-ended professional assignments written by domain experts and graded against typically 15 to 40 weighted expert rubric checkpoints per task. Weighted rubric score in percent on the law part of the XpertBench-Gold subset, judged by Gemini 2.5 Pro; higher is better.

Metric: Weighted rubric score (%) on the XpertBench-Gold subset (245 expert-written open-ended professional tasks, 15-40 weighted binary checkpoints each); ShotJudge: Gemini 2.5 Pro judges every checkpoint with an expert-graded GPT-5 response as one-shot exemplar; law tasks; higher is better. Source: arxiv.org. Saturation forecast: Around August 2027. 14 models tracked. Updated 7 Oct 2026.

Top models

#ModelSettingScoreOverall rank
1Claude Opus 4.6Thinking65.54#113
2Doubao-2.0-pro65.06#149
3GPT-5.4High64.79#135
4Claude Sonnet 4.5Thinking58.7#261
5Kimi K2 (Thinking)58
6GPT-5.2High56.88#185
7GPT-5High54.7#178
8Gemini 3.1 Pro (Preview)53.88#99
9DeepSeek V3.1 TerminusThinking51.9#437
10Gemini 3 Flash50.22#175
11Doubao-Seed-1.6Thinking49.4
12Gemini 3 Pro48.59#145
13Gemini 2.5 Pro47.3#276
14GLM-4.647.2#498

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

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