← All benchmarks

Humanity's Last Exam with tools (LFM3 report, cohort unspecified)

General QA

Humanity's Last Exam answered with tools as run by Liquid AI for its LFM3 launch table; the card names neither the tools, the judge nor the question cohort. All six reported values fall on a 0.1-point grid, which a single run over the 2,500-question set or its 2,158-question text-only subset cannot produce, so the observations are kept apart from the full-set with-tools benchmark.

Top models (higher is better)