AIThe Decoderabout 1 hour ago
New benchmark shows AI models fail visual perception
1 min read
Moonshot AI's PerceptionBench tests multimodal AI models' visual perception separately from reasoning. No frontier model reaches 60 percent accuracy, with GPT-5.6 Sol leading narrowly. Many reasoning errors actually occur during image reading, not later logic stages.
Level
Hype check
Tap to vote and see what everyone thinks.
Summary by ByteBrief
