NVIDIA AI Releases Nemotron 3 Embed: An Open Embedding Collection Whose 8B Checkpoint Ranks #1 on RTEB
NVIDIA has released Nemotron 3 Embed, a collection of three open embedding model checkpoints. The 8B parameter model achieves the top rank on the RTEB benchmark, demonstrating strong retrieval performance.
- NVIDIA's Nemotron-3-Embed-8B model sets a new standard, ranking first on the RTEB benchmark for retrieval performance.
- The release includes multiple model sizes and a quantized version (NVFP4) for efficient inference, maintaining high accuracy.
- All Nemotron 3 Embed models support a 32,768 token context window, enabling processing of longer documents.
- The models are open checkpoints, making advanced embedding capabilities accessible to the developer community.
NVIDIA AI has launched Nemotron 3 Embed, a new suite of open embedding models. The collection includes three checkpoints: Nemotron-3-Embed-8B-BF16, Nemotron-3-Embed-1B-BF16, and Nemotron-3-Embed-1B-NVFP4. The 8B model has achieved the highest ranking on the Retrieval Task Evaluation Benchmark (RTEB), scoring 78.46 on the NDCG@10 metric. The 1B models were developed using neural architecture search pruning and distillation techniques from the 8B model. Notably, the NVFP4 version offers nearly identical retrieval accuracy to the BF16 version while doubling throughput on Blackwell hardware.
All models in the Nemotron 3 Embed collection support extended context lengths of 32,768 tokens, powered by the OpenMDW-1.1 framework. This release provides developers with high-performance, open-source embedding solutions optimized for retrieval tasks and efficient deployment.
Provides access to a top-tier, open-source embedding model for building advanced retrieval systems.
Enables more accurate and efficient information retrieval, search, and recommendation engines.
Highlights NVIDIA's continued innovation in foundational AI models and commitment to open-source contributions.
Advances the state of the art in AI embedding models for better information access.
- RTEB
- Retrieval Task Evaluation Benchmark, a standard for measuring the performance of information retrieval models.
- NDCG@10
- Normalized Discounted Cumulative Gain at 10, a metric measuring the ranking quality of retrieval results.
- BF16
- BFloat16, a 16-bit floating-point format offering a good balance between precision and memory efficiency for AI models.
- NVFP4
- A specific NVIDIA quantization format (likely 4-bit) designed for efficient inference with minimal accuracy loss.
- OpenMDW-1.1
- An open-source framework from NVIDIA for handling large context windows in models.
Chinese startup Moonshot AI unveils Kimi model it says rivals OpenAI, Anthropic - CNBC
China's Moonshot unveils world's largest open AI model, closing in on US rivals - Yahoo Finance
LLMKimi K3 achieves 3rd Place on ArtificalAnalysis, beating out Claude Opus 4.8
LLMMoonshot’s upcoming Kimi 3 is expected to close the gap with Anthropic’s Opus 4.8
NVIDIA supports AI development in Japan with Nemotron - Engineering.com
New Training for Using Artificial Intelligence at Work - PR Newswire
A new training program for using artificial intelligence at work has been announced. The program aims to help employees develop skills in AI applications.
Tired Of Building AI Agents? There's A Simpler Way To Work Smarter - Forbes
Forbes highlights a new method for building AI agents without the complexity of traditional approaches.
RoboticsA Humanoid Company Backed by Eric Trump Is Preparing Its Robots for War
Foundation Future Industries, a humanoid robotics company advised by Eric Trump, is reportedly exploring military applications for its robots, including "kinetic things," according to its CEO.
Honda, Acura add Google Gemini AI assistant in US models - Just Auto
Honda and Acura are integrating Google's Gemini AI assistant into select US models, upgrading the in-car voice experience with generative AI capabilities.
Artificial Intelligence, Adaptive Radiotherapy and Treatment Automation Transform Global Cancer Care - GlobeNewswire
Artificial intelligence and adaptive radiotherapy are transforming global cancer care. Treatment automation is also playing a key role in this transformation.
Xi pitches China as leader of new global AI order, challenging US dominance - Reuters
China's president Xi Jinping aims to establish China as a leader in the global AI order, challenging US dominance. This move is part of China's efforts to become a major player in the AI industry.