NVIDIA Nemotron 3 Embed: Open-Source Embedding Models, 8B Tops RTEB
Decision Brief
What changedNVIDIA released open-source Nemotron 3 Embed models (8B, 1B BF16, 1B NVFP4), with the 8B ranking first on RTEB (average NDCG@10=78.46).
Why it mattersThe NVFP4 quantized model retains >99% retrieval accuracy while doubling Blackwell throughput, drastically cutting inference costs for NVIDIA hardware users.
Who should careOpen-source model users, Inference / infra teams
Affected stackNVIDIA
Source confidenceMedium · Reliable media or first-hand reporting
NVIDIA released Nemotron 3 Embed on July 15-16, 2026, with three checkpoints: 8B BF16, 1B BF16, and 1B NVFP4. The 8B model tops RTEB with average NDCG@10 of 78.46. The 1B BF16 model is pruned via ModelOpt NAS and distilled from the 8B teacher using cosine similarity and MSE. The 1B NVFP4 model achieves up to 2x throughput on Blackwell while preserving >99% retrieval accuracy of the BF16 version. All models support 32,768-token input and are released under the OpenMDW-1.1 license.
Summary basis: official / RSS sourceCompiled from the source scope noted above; the original remains authoritative.
Sources
- MarkTechPost
Fast research-paper and ML tooling summaries, useful for infra and agent updates.
- MarkTechPost
留言
登入后即可留言，和其他 builder 交换实测心得。
还没有留言，抢头香。