Benchmark RadarAI BENCHMARK PROFILE
ChronoPhyBench
ChronoPhyBench evaluates multimodal LLMs on chronological physical dynamics reasoning via next-state prediction and VQA, using video frames and captions for single-image selection and multi-frame sorting.
- Released
- 2026-06-06
- Readiness
- Paper only
- Primary field
- Cybersecurity
Why it matters
Addresses the gap of distinguishing true multimodal reasoning from language-prior exploitation, providing a robust framework to measure physical reasoning and hallucination rates in models.
Motivation
Recent advancements in Multimodal Large Language Models (MLLMs) have demonstrated remarkable proficiency in open-world reasoning and understanding.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.