Humanity’s Sixth Sense reportedly finds a wide human-model gap

Elorian AI and Scale AI have introduced Humanity’s Sixth Sense, a benchmark aimed at the intuitive visual reasoning people use in everyday situations. Scale Labs says the test contains 522 open-ended tasks spanning images and video, with questions covering spatial reasoning, causal reasoning and social understanding.

The organizers report a substantial gap between people and current models. Scale Labs says humans scored 93.1%, while GPT-6 Astra led the tested models at 53.6%. The median model scored 30.9%.

Elorian AI describes the benchmark as testing common-sense questions that people find easy, such as whether an apple will fit inside a lunchbox. Its announcement links to a blog post and a dataset page.

The scores are organizer-reported. The public announcements do not include the benchmark’s construction process, grading method, full model list, confidence intervals or contamination analysis, so the results cannot be independently evaluated from this evidence alone.