LaborLawBench: leaderboard
LaborLawBench (Chinese Labor Law Large Language Model Benchmark, Lan et al., 2026) spans twelve Chinese labor-law tasks: statute recitation, single- and multiple-choice exam questions, case-type and welfare-compensation prediction, named entity recognition, claim and dispute-focus mining, statute prediction and case analysis. The score is the unweighted mean of the twelve task scores (accuracy, F1, soft F1, ROUGE-L or a GPT-o1 judge score, each on a 0-1 scale), in percent; refused or unparsable answers score zero, and higher is better. The authors' LaborLawLLM is fine-tuned on labor-law instructions written in the benchmark's own task formats.
Metric: Mean Task Score (%; unweighted mean of 12 labor-law task scores on a 0-1 scale). Source: arxiv.org. Saturation forecast: Estimated already saturated. 26 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | DeepSeek V3 (0324) | 61 | #704 |
| 2 | GPT-4o | 57 | #741 |
| 3 | Moonshot v1 8K | 54 | #1540 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=laborlawbench · How It Works · Data refreshed daily, snapshot 2026-11-12.