To request an evaluation, submit a publicly available Hugging Face retrieval model through our evaluation request form: https://t.co/yGvVSxnrC3 The full technical report with methodology and findings is available here: https://t.co/G0AKX2N8r5
AI Weekly's analysis
→
- Q2D-Web pairs 190 million web documents with 70,000 agentic search queries across 10 languages, reformulated from real user queries.
- Thirteen retrievers spanning lexical, dense, and late-interaction families are scored against three relevance sets, including agent citations and production rankings.
- Retriever rankings hold across judgment sets but shift with topical domain, query language, and query type.
Read full analysis →
