LegalCiteBench - Case Verification: 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 case verification the model checks whether a cited case supports a stated principle and corrects it if not; judge score from 0 to 100, higher is better.

Metric: Cat4-2 judge score (0-100) for verifying whether a cited case supports a stated legal principle and supplying the correct case when it does not (6,377 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
1Mistral 7B96.07#7794
2DeepSeek-R1-Distill-Qwen-7B91.44#8055
3DeepSeek R1 Distill Qwen 1.5B90.81#9215
4Gemini 2.5 Flash89.88#463
5Qwen 3 14B89.58#1269
6Qwen 2.5 7B89.32#2929
7Llama 3.1 70B89#1499
8GPT-5 Mini88.28#364
9Gemma 3 12B85.91#2059
10SaulLM-54B82.14
11O4 Mini80.12#368
12DeepSeek V3.178.16#511
13Llama 3.2 3B78.1#7877
14Qwen 3 30B A3B78.04#1156
15Llama 3.1 8B76.31#5196

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

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