Benchmark RadarAI BENCHMARK PROFILE
StylisticBias
StylisticBias is a benchmark of 25,000 images with controlled single-attribute variations to evaluate attribute-level social bias in multimodal LLMs. It measures how visual cues shift model judgments in 25 scenarios.
- Released
- 2026-06-18
- Readiness
- Runnable
- Primary field
- General AI
Why it matters
Social bias in multimodal models is often confounded by identity. StylisticBias isolates visual cues, enabling targeted bias diagnosis and mitigation.
Motivation
Multimodal large language models (MLLMs) are increasingly deployed in personally and societally consequential settings, yet the visual cues that shape how these models judge people remain poorly understood.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.