LongJud-Bench - Legal Reasoning (First Instance): leaderboard

Metric: Legal Reasoning Score (%; Claude Sonnet 4.6 judge, 0-10 per trial phase on legal-reasoning sufficiency in the first-instance trial statements, normalized). Source: arxiv.org. Saturation forecast: Around December 2026. 6 models tracked.

Top models

#ModelScore
1GPT-5.257
2Kimi K2.556
3Qwen 3.5 Plus53
4DeepSeek V4 Flash52
5GLM-4.748
6Qwen 3.5 Flash45

Interactive version: theaggregate.ai/benchmark?slug=longjud-bench-legal-reasoning-first-instance · How It Works · Data refreshed daily, snapshot 2026-09-25.