WANDR

A new benchmark called WANDR (Wide ANd Deep Research) has been introduced to evaluate research agents on 500 realistic data-collection tasks involving entity discovery and evidence-backed claims, using a reference-free grader that re-fetches cited pages. The benchmark reports precision, recall, and F1 scores over hierarchical task structures, with Perplexity Search as Code (GPT-5.5) ranking first at 36.30.

WANDR Wide ANd Deep Research evaluates research agents on 500 realistic, high-volume data-collection tasks that require broad entity discovery, systematic enrichment, and evidence-backed claims. Its reference-free grader re-fetches cited pages and reports soft and hard precision, recall, and F1 over hierarchical task structures. Category: Agentic. Imported rows: 6. Top imported result: Perplexity Search as Code GPT-5.5, high , rank 1, 36.30.