Benchmark RadarAI BENCHMARK PROFILE
MoHallBench
MoHallBench evaluates motion hallucination in video LLMs with 11,306 video clips and 40,493 QA pairs, covering three hallucination sources and multiple choice settings.
- Released
- 2026-07-01
- Readiness
- Paper only
- Primary field
- General AI
Why it matters
Targets a specific video understanding failure mode, offering metrics to reduce affirmation bias and revealing gaps in action recognition vs. hallucination resistance.
Motivation
Video Large Language Models (VideoLLMs) have shown strong progress in video understanding, yet they still suffer from hallucinations that are inconsistent with visual evidence.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.