LegalCiteBench - Citation Retrieval: 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 retrieval the model lists the authorities an opinion relied on for a legal research question; citation-level F1 from 0 to 100, higher is better.

Metric: Cat1 citation-level F1 (0-100) between the cited authorities the model lists for an opinion-derived legal research question and the reporter citations of the source opinion (4,899 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: Around 2030. 21 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Claude Sonnet 4.56.8#267
2O4 Mini5.73#372
3DeepSeek V3.15.31#511
4Gemini 2.5 Flash5.2#466
5Claude Haiku 4.54.42#587
6Llama 3.1 70B3.82#1509
7SaulLM-54B3.77
8Qwen 3 30B A3B3.65#1165
9Phi-43.59#1951
10Qwen 3 4B3.33#2379
11Mistral 7B2.85#7811
12Qwen 3 14B2.54#1284
13GPT-4o Mini2.4#1680
14Gemma 3 12B1.92#2091
15GPT-5 Mini1.81#373

No result here: #2 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=legalcitebench-citation-retrieval · How It Works · Data refreshed daily, snapshot 2026-11-22.