NVIDIA Releases New Retrieval Embedding Model
kimmonismus · x · 2026-07-17
NVIDIA launched a new embedding model, Nemotron-3-Embed-8B, claiming it achieves an average NDCG@10 of 78.46 on RTEB and 75.45 on MMTEB Retrieval.
Official tests integrated it into a search agent powered by Nemotron 3 Ultra. Results show that more accurate retrieval surfaces relevant evidence earlier, reducing repeated searches, context checks, and reasoning turns, which lowers downstream token costs. Alongside the 8B version, NVIDIA released two 1B variants; the NVFP4 version optimized for Blackwell purportedly achieves up to 2× BF16 throughput while retaining over 99% retrieval quality. All three support 32K context, multilingual and code retrieval, enterprise RAG, and agent memory, with open weights, data, and recipes.
Related event: NVIDIA launches Nemotron-3-Embed retrieval models(11 posts)→
More from Infra
- oMLX 0.5.2 adds Mac menu-bar stats, low-bit decode kernels, and faster downloads — awnihannun · 2026-07-22
- Strangeworks launches Aura to turn enterprise ops into production optimization systems — whurley · 2026-07-22
- Graph workload 854.graph500 enters SPEC CPU 2026 as a new CPU benchmark — Prof_DavidBader · 2026-07-22
- Hybrid and local inference are emerging as a response to AI energy and token costs — dmitry140 · 2026-07-22
- NVIDIA details Vera CPU with 2x performance claims and a 22,000-core rack — ryanshrout · 2026-07-22
- NVIDIA says Vera Rubin NVL72 delivers 10x more tokens per megawatt than Blackwell — nvidia · 2026-07-22