Benchmark RadarAI BENCHMARK PROFILE
PerceptionBench
PerceptionBench evaluates atomic visual perception in MLLMs with 3,000 verified questions isolating ten perceptual capabilities, based on an error taxonomy from 42 benchmarks.
- Released
- 2026-07-27
- Readiness
- Paper only
- Primary field
- General AI
Why it matters
Addresses the need for a capability-level standard to diagnose visual perception boundaries, showing that current MLLMs remain below 60% accuracy.
Motivation
We introduce PerceptionBench, a benchmark specifically designed to evaluate the atomic visual perception capabilities of Multimodal Large Language Models (MLLMs).
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.