Benchmark
ATLAS
Test how accurately and completely agents research entities and assemble sourced answers.
Explore the original evidence
Explore the original evaluation, its methods, and reported findings.
Open evalCoverage
The October preview reports 547 search-intensive tasks with verified answer keys.
Methodology
Score entity discovery, complete rows, and individual cells with separate precision-and-recall measures. Compare configurations with their stated model, harness, and per-task costs.