Verified, then logged.
Sun 22 Nov 2026 · 00:24 KST / 15:24 UTC한국어 | ENSearch /

Breaking01 / AI

ChJGBench tests AI generation of Chinese court judgments

Researchers reported that legal models' strengths over general-purpose models varied by evaluation category.

An evaluator seen from behind reviews AI-generated judgments and comparison charts across multiple monitors in a legal research office, with all screen content blurred and unreadable. AI-generated image

Researchers announced ChJGBench on 13 Nov, an evaluation benchmark built from 3,462 real judgments for AI models generating Chinese court judgments.

According to the paper's abstract, the benchmark covers civil, criminal and administrative cases. It evaluates the generation of judicial reasoning and the operative section, which sets out the court's final decision. Its framework uses 13 metric types tailored to case categories.

The researchers identified a need to assess whether models apply statutes accurately and resolve issues that determine outcomes. These include claims, charges, penalties and monetary obligations.

In their published experimental findings, the researchers reported no consistent advantage for legal models over general-purpose models. Strengths varied by evaluation category, with generally better performance on criminal cases and continuing difficulty with civil and administrative cases.

The abstract described earlier benchmarks as focusing on isolated legal tasks or broad assessments.

The researchers did not disclose model-level scores or detailed definitions of each metric in the abstract.