Moonshot AI Releases PerceptionBench for Visual Perception

Moonshot AI launched PerceptionBench, a 3,000-question benchmark designed to isolate atomic visual perception from reasoning and knowledge errors in multimodal models. Results show leading models still struggle to surpass 60% accuracy.

2026-07-28 ~ 2026-07-29 · 2 related posts

• Moonshot releases PerceptionBench, a 3,000-question visual perception benchmark — zainhas · 2026-07-28
• PerceptionBench finds no frontier MLLM reaches 60% on atomic visual perception — moonshotai · 2026-07-29