LegalCiteBench - Citation Error Detection: leaderboard

LegalCiteBench (2026) tests whether LLMs, without retrieval, can recover and verify the case authorities cited in 1,000 U.S. judicial opinions from the Case Law Access Project. In citation error detection the model finds and corrects wrong citations in legal analysis paragraphs; judge score from 0 to 100, higher is better.

Metric: Cat3 judge score (0-100) for detecting citation errors in legal analysis paragraphs and supplying the correct citation (5,474 instances), closed-book (no retrieval) on instances derived from 1,000 U.S. judicial opinions of the Case Law Access Project, greedy decoding at temperature 0 with at most 400 generated tokens, GPT-4o-mini judge; higher is better. Source: arxiv.org. Saturation forecast: Not forecast. 21 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1SaulLM-54B75.59
2DeepSeek-R1-Distill-Qwen-7B58.71#8055
3DeepSeek R1 Distill Qwen 1.5B58.11#9215
4GPT-5 Mini54.95#364
5Qwen 3 14B52.18#1269
6Qwen 3 4B47.12#2348
7Gemini 2.5 Flash46.93#463
8O4 Mini46.71#368
9Llama 3.1 8B45.32#5196
10Claude Sonnet 4.544.87#261
11Qwen 3 30B A3B43.55#1156
12Mistral 7B43.4#7794
13Phi-443.35#1931
14Claude Haiku 4.543.01#583
15GPT-4o Mini42.84#1667

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

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