Benchmark RadarAI BENCHMARK PROFILE
RuVerBench
RuVerBench evaluates LLM-as-a-judge reliability for rubric verification in agentic scenarios. It includes 2,458 instances across deep research and agentic coding, each with a model-generated output, a rubric, and a human-annotated label indicating rubric satisfaction.
- Released
- 2026-06-29
- Readiness
- Runnable
- Primary field
- General AI
Why it matters
Rubric-based scoring with LLM judges is common but under-validated, especially for agentic outputs. RuVerBench provides a reusable benchmark to compare judge models and strategies, enabling decisions on model selection and scoring protocol.
Motivation
Rubric-based scoring has become a widely used paradigm in model evaluation, typically with LLM-as-a-Judge (LaaJ) for rubric scoring.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.