Expanding the world of open models, NVIDIA today released new open models, data, and tools to advance AI across all industries.
These models span the NVIDIA Nemotron family for agent AI, the NVIDIA Cosmos platform for physical AI, the new NVIDIA Alpamayo family for autonomous vehicle development, NVIDIA Isaac GR00T for robotics, and NVIDIA Clara for biomedical, giving companies the tools to develop real-world AI systems.
NVIDIA contributes to open source training frameworks and one of the world's largest collections of open multimodal data, including 10 trillion language training tokens, 500,000 robot trajectories, 455,000 protein structures, and 100 terabytes of vehicle sensor data. It is a diverse open resource of unprecedented scale to accelerate innovation in languages, robotics, scientific research, and self-driving cars.
Leading technology companies including Bosch, CodeRabbit, CrowdStrike, Cohesity, Fortinet, Franka Robotics, Humanoid, Palantir, Salesforce, ServiceNow, Hitachi, and Uber are adopting and building on NVIDIA's open model technology.
NVIDIA Nemotron brings voice, multimodal intelligence, and safety to AI agents
Building on the recently released NVIDIA Nemotron 3 family of open models and data, NVIDIA is releasing Nemotron models for voice, multimodal search augmented generation (RAG), and safety.
- Nemotron's speech is comprised of leaderboard-topping open models, including a new ASR model that delivers real-time, low-latency speech recognition for live captioning and voice AI applications. Daily and modal benchmarks show that this model achieves 10x faster performance than other models in its class.
- Nemotron RAG consists of a novel embedding and re-ranking vision language model (VLM) that provides highly accurate multilingual and multimodal data insights to enhance document retrieval and information retrieval.
- Safety of Nemotron Models that enhance the safety and reliability of AI applications now include the Llama Nemotron Content Safety model, which features expanded language support, and Nemotron PII, which detects sensitive data with high accuracy.
Bosch uses Nemotron Speech to enable drivers to interact with their vehicles. ServiceNow trains the Aprilel model family on open datasets, such as Nemotron, for cost-effective multimodal performance.
Cadence and IBM are piloting the NVIDIA Nemotron RAG model to improve search and inference across complex technical documents.
CrowdStrike, Cohesity, and Fortinet have adopted the NVIDIA Nemotron Safety model to strengthen the reliability of their AI applications.
Palantir is integrating Nemotron models into its ontology framework, building a first-of-its-kind integrated technology stack for specialized AI agents. CodeRabbit uses Nemotron models to power and extend AI code reviews, increasing speed and cost efficiency while maintaining high review accuracy.
NVIDIA is also releasing open source datasets, training resources, and blueprints to developers, including datasets and training code for the Llama Embed Nemotron 8B model featured on the MMTEB leaderboard. This is in addition to an updated LLM router that shows developers how to automatically direct AI requests to the best model for the job, and a dataset used to build the new Nemotron Speech ASR model.
New models for all types of physics AI and robots
Developing physical AI for robots and autonomous systems requires large and diverse datasets and models that can perceive, reason, and operate in complex real-world environments. At Hugging Face, robotics is the fastest growing segment, with NVIDIA's open robotics models and datasets leading platform downloads.
NVIDIA is releasing the NVIDIA Cosmos Open World Foundation Model, which brings human-like reasoning and world generation to accelerate the development and validation of physical AI.
NVIDIA also released open models and blueprints for each physical AI embodiment built on Cosmos.
- Isaac GR00T N1.6 is an open reasoning Vision Language Action (VLA) model built specifically for humanoid robots to unlock whole-body control and improve reasoning and context understanding using NVIDIA Cosmos Reason.
- NVIDIA Blueprint for Video Search and SummarizationA reference workflow for building vision AI agents, part of the NVIDIA Metropolis platform, that can analyze large volumes of recorded and live video to improve operational efficiency and public safety.
Salesforce, Milestone, Hitachi, Uber, VAST Data, and Encord use Cosmos Reason for their traffic and workplace productivity AI agents. Franka Robotics, Humanoid, and NEURA Robotics use Isaac GR00T to simulate, train, and validate new robot behaviors before scaling up to production.
NVIDIA Alpamayo for inference-based self-driving cars
Developing safe, scalable autonomous driving relies on AI that can perceive, reason, and act on complex real-world environments and scenarios, with development workflows that support rapid training, testing, and improvement at scale.
NVIDIA releases NVIDIA Alpamayo, a new family of open models, simulation tools, and large-scale datasets to accelerate inference-based self-driving vehicle development. This includes:
- Alpamayo 1the first open, large-scale inference VLA model for autonomous vehicles (AVs) that enables vehicles to understand their environment and explain their behavior.
- alpacimis an open-source simulation framework that enables closed-loop training and evaluation of inference-based AV models across a variety of environments and edge cases.
NVIDIA has also released Physics AI Open DatasetIt includes over 1,700 hours of driving data collected across a wide range of regions and conditions, covering rare and complex real-world edge cases essential to advancing inference architectures.
NVIDIA Clara for Healthcare and Life Sciences
To reduce costs and deliver treatments faster, NVIDIA launches new Clara AI models that bridge the gap between digital discovery and real-world medicine.
These models that help researchers design treatments that are safer, more effective, and easier to manufacture include:
- la proteina It enables the design of proteins with great atomic precision for research and drug candidate development, giving scientists new tools to study diseases previously thought to be untreatable.
- ReaSyn v2 Incorporating manufacturing blueprints into the discovery process ensures that AI-designed drug synthesis is practical.
- celtic provides highly accurate computational safety testing early in development by predicting how potential drugs will interact with the human body.
- RNA Pro Unlock the potential for personalized medicine by predicting the complex 3D shapes of RNA molecules.
Additionally, the NVIDIA dataset of 455,000 synthetic protein structures helps AI researchers build more accurate AI models.
Explore NVIDIA open models and technologies
NVIDIA's open models, data, and frameworks will be available on GitHub and Hugging Face, as well as a variety of cloud, inference, and AI infrastructure platforms, and build.nvidia.com, giving developers flexible access to support resources.
Many of these models are also available as NVIDIA NIM microservices, which can be securely and scalably deployed on NVIDIA-accelerated infrastructure from the edge to the cloud.
Watch and learn more NVIDIA Live at CES.
