LexIssue - Legal Issue Generation (Knowledge-Base RAG): leaderboard
Legal issue generation in LexIssue (Xie et al., 2026): from synthesized claim and defence forms of 430 real Chinese first-instance civil cases, the model states the disputed legal issues in free form. Case-level weighted F1 in percent against 1,303 expert-annotated issues after alignment by a Qwen-3.5-27B judge, with the top five entries of the LexIssue legal issue knowledge base for the cause of action added to the prompt; higher is better.
Metric: Weighted F1 (%; legal issue generation: free-form disputed-issue descriptions matched to the 1,303 reference issues; predicted issues are aligned to the expert reference issues by a two-stage Qwen-3.5-27B judge (semantic retrieval, then rubric-based equivalence verification); mean case-level weighted F1 over the 430 cases; retrieval-augmented setting: the top 5 entries of the authors' 441-entry legal issue knowledge base for the case's cause of action, reranked by Qwen3-Reranker-8B, are added to the prompt). Source: arxiv.org. Saturation forecast: Around February 2027. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Flash | 71.5 | #159 |
| 2 | GPT-5.2 | 63.4 | #172 |
| 3 | DeepSeek V4 Flash | 62.8 | #254 |
| 4 | Qwen 3 8B | 49.5 | #1838 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=lexissue-legal-issue-generation-knowledge-base-rag · How It Works · Data refreshed daily, snapshot 2026-11-12.