GPT-6-astra was the top performing model in the test but reached only 53.6%. Co-creator Andrew Dai noted that the benchmark exposes a blind spot where models that can write professional code fail to identify basic visual facts, such as whether a bookshelf has room for an additional book.
Wednesday, October 21, 2026
Story
GPT-6-astra Scores 53.6% on New Humanity's Sixth Sense Benchmark
AI13 days ago
Scale AI and Elorian AI have released a multimodal evaluation measuring the ability of artificial intelligence to navigate images and video with everyday human intuition. The project, named Humanity's Sixth Sense (HSS), uses 522 open ended tasks to test spatial, causal, and social reasoning. In initial results, the median AI model scored 30.9%, while humans averaged 93.1%.
Looking for more context or analysis?
Tilt chat provides reliable, tailored insights, alerts, and automations.