Intel / post

NVIDIA's Nemotron 3 Embed Tops a New Web Retrieval Benchmark

Source: x.com
Date: Published Sep 10, 2026

Context

NVIDIA's Nemotron 3 Embed 8B model took the top combined score for nDCG@10, a measure of how highly a system ranks the single most relevant document among its top 10 results, on Q2D-Web, a retrieval Perplexity built specifically for agentic search. The test set covers 190 million web documents and nearly 70,000 search queries that were reformulated by an rather than typed by a person, across 10 languages.

That distinction matters because agents phrase queries differently than people do, often longer and more specific, so a benchmark built on agent-generated queries tests something closer to how retrieval models actually get used inside agent pipelines. NVIDIA's result gives engineers picking an model for agentic or web search a data point measured under those conditions rather than on a benchmark built from human search logs.

Terms in this piece · Glossary

benchmark — A standard public test set for comparing AI models — the shared scoreboards behind every "model X beats model Y" claim.
AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
embedding — A list of numbers representing a piece of text's meaning, so that similar meanings end up numerically close and can be searched.
RAG — Retrieval-augmented generation — fetching relevant documents first and pasting them into the model's context so it answers from your data instead of memory.

Read the source x.com