About

A practical map, not a final authority.

FronteraEval began during Miguel Guerrero’s Cambridge ERA research fellowship. Finding a current, comprehensive, methodologically careful, and simple point of entry into frontier-AI evaluations was harder than it should have been.

FronteraEval is an attempt to reduce that friction. It does not claim to be complete, neutral, or definitive. Upstream metadata changes; classifications involve judgement; and most of the 314 catalogue records are source-linked discovery entries rather than independent reviews.

30 records currently contain bounded documentary methodological assessments. These are not experimental replications. Important claims should still be checked against the original paper, code, task version, model-system configuration, scaffold, and evaluation date.

Corrections and missing evaluations are welcome through GitHub issues.