Benchmark RadarAI BENCHMARK PROFILE
CompanionBench
CompanionBench is an interactive bilingual benchmark for AI emotional companionship, grounding scenarios and a user simulator in de-identified real-world data. It evaluates ten capabilities derived from 25 theories, using a rubric and deterministic disclosure measure.
- Released
- 2026-08-03
- Readiness
- Paper only
- Primary field
- General AI
Why it matters
LLM companions are deployed at scale but poorly evaluated. CompanionBench provides a reproducible, theory-anchored benchmark with real-world grounding, offering granular capability assessment and addressing judge biases.
Motivation
LLM companions are deployed at scale in personally consequential settings, yet poorly evaluated.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.