← All benchmarks

Humanity's Last Exam - full verified 1,811-item set

General QA

Accuracy on the full revised and verified HLE set: 668 verified original questions and 1,143 revised questions, excluding 689 original questions classified as uncertain. Google identifies this exact cohort for the Gemini 3.7 and 3.8 Flash evaluations, citing arXiv:2602.13964. The verified cohort is distinct from the original 2,500-item HLE and text-only variants.

Top models (higher is better)