LegalCiteBench - Citation Completion: 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 completion the model supplies the source-opinion authorities missing from a partial citation list; citation-level F1 from 0 to 100, higher is better.

Metric: Cat2 citation-level F1 (0-100) of the remaining source-opinion authorities the model supplies given a partial citation set (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 2037. 21 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Claude Sonnet 4.56.35#261
2Qwen 3 14B6.34#1269
3DeepSeek V3.15.15#511
4DeepSeek R1 Distill Qwen 1.5B4.54#9215
5Qwen 3 30B A3B4.34#1156
6Llama 3.1 70B4.34#1499
7O4 Mini4.06#368
8SaulLM-54B4.06
9Claude Haiku 4.53.87#583
10Gemini 2.5 Flash3.66#463
11Qwen 2.5 7B3.5#2929
12DeepSeek-R1-Distill-Qwen-7B3.49#8055
13Qwen 3 4B3.37#2348
14Phi-43.16#1931
15Mistral 7B2.56#7794

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-completion · How It Works · Data refreshed daily, snapshot 2026-11-21.