Humanity's Last Exam

Humanity’s Last Exam is a set of questions written by specialists across dozens of subjects, chosen because they are hard enough that top models get most of them wrong. It exists because the older benchmarks got saturated: once every leading model scores above 90 percent, the test stops telling you anything.

Scores are still low by the standards of other benchmarks, in the 30 percent range for strong models in mid-2026, which is the point. It measures the gap that is left rather than the ground already covered. Like any single benchmark, it captures one narrow slice of capability and should not be read as a general intelligence score.