LexRubric: leaderboard

Metric: Overall score rate (%): points earned from expert-written atomic rubric items (weighted -10 to +10; negative items subtract) divided by the maximum positive score, averaged over all 649 instances (473 legal consultation, 176 judicial examination); a Qwen3.6-27B judge checks each item, temperature 0.6, 16k output tokens; can fall below 0; higher is better. Source: arxiv.org. Saturation forecast: Around September 2027. 18 models tracked.

Top models

#ModelScore
1Kimi K2.675.21
2Qwen 3.6 Max Preview75.12
3Qwen 3 Max74.35
4Kimi K2.573.4
5GLM-5.169.96
6Claude Sonnet 4.669.18
7Qwen 3.5 397B A17B68.03
8GLM-563.28
9DeepSeek V4 Flash63.19
10DeepSeek V4 Pro62.22
11DeepSeek V3.260.03
12DeepSeek R159.49
13GPT-5.256.43

Interactive version: theaggregate.ai/benchmark?slug=lexrubric · How It Works · Data refreshed daily, snapshot 2026-09-29.