NVIDIA AI Releases Nemotron 3 Embed: An Open Embedding Collection Whose 8B Checkpoint Ranks #1 on RTEB

Short summary

NVIDIA released Nemotron 3 Embed, a collection of three open embedding checkpoints (8B, 1B BF16, 1B NVFP4). The 8B model ranks #1 on RTEB with 78.46 average NDCG@10, while the 1B was produced via NAS pruning and distillation from the 8B teacher. The NVFP4 variant retains 99%+ of BF16 retrieval accuracy at up to 2x Blackwell throughput, and all three support 32,768-token inputs under OpenMDW-1.1.

• •Nemotron-3-Embed-8B-BF16 ranks #1 on RTEB at 78.46 NDCG@10
• •1B variants derived via ModelOpt NAS pruning and COS+MSE distillation
• •NVFP4 checkpoint retains 99%+ BF16 accuracy at up to 2x Blackwell throughput

Generated with AI, which can make mistakes.