Released September 10, 2026 alongside arXiv:2609.08887, Q2D-Web (Query2Doc-Web) extends Perplexity's earlier Q2D embedding benchmark to web scale. Queries are reformulated from 23,000 PII-free production searches over nine months across ten languages (English ~65.8%). Three fixed judgment sets — agent citations, production web rankings, and a combined LLM-judged union — average about 99.6 positive labels per query in the combined view. A public Hugging Face leaderboard accepts open embedder submissions; the paper documents subcorpus sampling to cut full-corpus eval cost.