Nemotron Embed
by NVIDIA · Model · Language & code models
Ranked first at the 1B scale on the retrieval leaderboard, scoring 72.4% (RTEB), a 27% error reduction over its predecessor
Evidence
- Aug 4NVIDIA chart shows it outperforming same-size embedders on four retrieval benchmarks; scores 85 on LongEmbed at 32K context
- Aug 4In a 3-question multi-hop agent test, needed 11 searches and 8,734 tokens versus 15 searches and 12,688 for Nomic: 31% fewer tokens
- Aug 4Multilingual support is good but narrower than hoped; reviewer wants low-resource regional languages covered
- Jul 19Embedding family (1B and 8B) beating similar-size Qwen 3 and Gemma embedders on retrieval, multilingual; 8B is 16 GB, 1B is 2.3 GB