Scale AI Labs introduced Humanity’s Sixth Sense, a benchmark for intuitive visual reasoning. It tests models on implicit temporal, spatial, social and abstract cues humans infer at a glance. Humans score 93.1% accuracy on the tasks. The best model, GPT-6-astra, scores 53.6% even with maximum reasoning effort. Agentic setups with dynamic visual manipulation narrow but do not close the gap. The benchmark covers diverse image and video inputs with a structured taxonomy. It highlights a capability that current scaling approaches have not addressed.

• 93.1% human accuracy
• 53.6% GPT-6-astra accuracy