<AI>Devspace

Best multimodal models still can't crack 50 percent on basic visual entity recognition

2/8/2026, 1:19:04 PM · The Decoder