NVIDIA Launches Nemotron-3-Embed Series
rohanpaul_ai · x · 2026-07-17
NVIDIA released the Nemotron-3-Embed series, a set of open, commercially available embedding models designed for agentic retrieval, code retrieval, and agent memory.
Key Highlights
- Nemotron-3-Embed-8B-BF16 ranks first on RTEB with a score of 78.5%
- Nemotron-3-Embed-1B-BF16 scores 72.4%, reducing error rates by 27% compared to the previous generation
- Two additional 1B variants are available: one optimized for cost/latency, and a Blackwell-optimized 1B-NVFP4 version claiming up to 2x throughput improvement while retaining 99%+ BF16 accuracy
- All three models support 32K context inputs, making them suitable for long documents, code, and agent history
Background
The original post emphasizes that failed retrieval forces agents into repeated searches, lengthens reasoning, and increases token costs; better retrieval directly reduces search frequency and overall expenses.
Related event: NVIDIA launches Nemotron-3-Embed retrieval models(11 posts)→
More from Infra
- Encrypted GPT-2 speedrun challenges FHE inference: best is 3.031 s/token, goal sub-1s — tensorqt · 2026-10-27
- Open-source Latch compiles PyTorch into training megakernels, 2.75x faster on H100 — sguada · 2026-10-27
- Stanford launches GPU-centric supercomputer Marlowe for data science and AI research — StanfordHAI · 2026-10-27
- NVIDIA Claims HGX Rubin NVL8 Delivers 10x More Tokens per Megawatt Than B200 — thursdai_pod · 2026-10-27
- Together lets multiple LoRA post-training experiments share the same GPU capacity — togethercompute · 2026-10-27
- Commoditization dilemma: either human labor or GPUs become the commodity — _AustinCalvert_ · 2026-10-27