LexAgentHallu - LawThinker - L2 Hallucination Density: leaderboard

LexAgentHallu evaluates hallucinations in multi-step legal agents. L1 covers substantive legal knowledge and L2 procedural reasoning. Hallucination frequency is the fraction of rollouts with at least one flagged error; density is the mean fraction of applicable error subclasses triggered. Frameworks are separate protocols; the LRAS fine-tuned models retain their own identities. This series reports LexAgentHallu - LawThinker - L2 Hallucination Density from 2609.09754v1, S4.T2: Hallucination Density (%). Lower is better.

Metric: Hallucination Density (%). Source: arxiv.org. Saturation forecast: Around February 2027. 5 models tracked.

Top models

#ModelScoreOverall rank
1GPT-5.417.2#125
2Gemini 3.1 Pro (Preview)23.3#93
3Qwen 3.6 27B32.7#345
4Qwen 3.5 27B37.5#368
5Qwen 3.5 9B47#862

No result here: #4 Claude Opus 5.5, #5 GPT-6 Astra, #9 GPT-6.1 Sol.

Interactive version: theaggregate.ai/benchmark?slug=lexagenthallu-lawthinker-l2-hallucination-density · How It Works · Data refreshed daily, snapshot 2026-11-08.