Benchmark RadarAI BENCHMARK PROFILE
SafeBuild-Bench
A benchmark for evaluating multimodal large language models on construction safety hazard identification and description, with 3,314 expert-verified task instances from over 3,000 images.
- Released
- 2026-07-29
- Readiness
- Runnable
- Primary field
- General AI
Why it matters
Construction-safety models must handle realistic temporal and site variation; this benchmark provides a standard evaluation for hazard recognition and description, useful for deployment risk assessment.
Motivation
Construction-safety models must handle concrete deployment risks, such as a worker standing near a scaffold edge without guardrails, rather than only recognize common objects in curated images.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.