NVIDIA Unveils New Open Models, Data and Tools to Advance AI Across Every Industry

Source: NVIDIA AI Blog

Source: NVIDIA AI Blog

NVIDIA Expands the Open‑Model Universe

NVIDIA today announced a suite of new open models, data, and tools designed to accelerate AI adoption across every industry.

Open Models

These models give companies the building blocks they need to create real‑world AI systems.

Open‑Source Resources & Data

• Training frameworks – Fully open‑source and ready for customization.
• Multimodal data collection – One of the world’s largest, including:
  • 10 trillion language training tokens
  • 500,000 robotics trajectories
  • 455,000 protein structures
  • 100 TB of vehicle sensor data

This unprecedented scale of diverse, open resources speeds innovation in language models, robotics, scientific research, and autonomous vehicles.

Early Adopters

Leading technology companies are already building on NVIDIA’s open‑model ecosystem, including:

• Bosch
• CodeRabbit
• CrowdStrike
• Cohesity
• Fortinet
• Franka Robotics
• Humanoid
• Palantir
• Salesforce
• ServiceNow
• Hitachi
• Uber

---

For more details, explore the linked resources above.

NVIDIA Nemotron Brings Speech, 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 speech, multimodal retrieval‑augmented generation (RAG), and safety.

---

Nemotron Speech

• Model: nemotron-speech-realtime-en-600m – a leaderboard‑topping open model for real‑time English speech.
• ASR variant: nemotron-speech-streaming-en-0.6b – low‑latency streaming speech‑to‑text for live captions and speech‑AI applications.
• Performance: Daily and Modal benchmarks show ≈10× faster inference than competing models in its class. See the full benchmark details here.

---

Nemotron RAG

• Collection: Nemotron RAG – a suite of vision‑language models for retrieval‑augmented generation.
• Embed model: llama-nemotron-embed-vl-1b-v2 – multilingual, multimodal embeddings.
• Rerank model: llama-nemotron-rerank-vl-1b-v2 – high‑accuracy reranking for document search and information retrieval.

---

Nemotron Safety

• Content‑safety model: Llama-3.1-Nemotron-Safety-Guard-8B-v3 – expanded language support for safe AI interactions.
• PII detection: gliner-PII – a GLiNER‑based model that detects personally identifiable information with high precision.

Early Adopters

Resources for Developers

Datasets & Training Code

• Embed Nemotron v1 dataset –
  https://huggingface.co/datasets/nvidia/embed-nemotron-dataset-v1
• Training code (Biencoder example) –
  https://github.com/NVIDIA-NeMo/Automodel/tree/main/examples/biencoder
  Used for the Llama Embed Nemotron 8B model: https://huggingface.co/nvidia/llama-embed-nemotron-8b (MMTEB leaderboard entry).
• Granary dataset –
  https://huggingface.co/datasets/nvidia/Granary
  Utilized to build the new Nemotron Speech ASR model.

Tools

• LLM Router – Updated version that shows developers how to automatically route AI requests to the most suitable model.
  https://build.nvidia.com/nvidia/llm-router

These open‑source assets, together with the Nemotron model families, provide a full stack for building speech‑enabled, multimodal, and safe AI agents.

New Models for Every Type of Physical AI and Robot

Developing physical AI for robots and autonomous systems requires large, diverse datasets and models that can perceive, reason, and act in complex, real‑world environments. On Hugging Face, robotics is the fastest‑growing segment, with NVIDIA’s open‑source robotics models and datasets leading the platform’s downloads — see the AI World story.

---

NVIDIA Cosmos Foundation Models

NVIDIA is releasing NVIDIA Cosmos open‑world foundation models that bring human‑like reasoning and world generation to accelerate physical‑AI development and validation.

---

Open Models & Blueprints Built on Cosmos

• Isaac GR00T N1.6 – An open reasoning vision‑language‑action (VLA) model purpose‑built for humanoid robots. It unlocks full‑body control and leverages Cosmos Reason for richer contextual understanding.
  → GitHub repo

• NVIDIA Blueprint for Video Search & Summarization – Part of the NVIDIA Metropolis platform, this reference workflow enables vision‑AI agents to analyze massive volumes of recorded and live video, boosting operational efficiency and public safety.
  → Blueprint page

---

Companies Leveraging Cosmos Reason

• Salesforce, Milestone, Hitachi, Uber, VAST Data, Encord – Using Cosmos Reason for traffic‑management and workplace‑productivity AI agents.
• Franka Robotics, Humanoid, NEURA Robotics – Deploying Isaac GR00T to simulate, train, and validate new robot behaviors before scaling to production.

“Cosmos Reason is enabling a new generation of AI agents that can understand and act in the physical world with unprecedented fidelity.” – NVIDIA AI Research

---

All links are current as of January 2026.

NVIDIA Alpamayo for Reasoning‑Based Autonomous Vehicles

Developing safe, scalable autonomous driving depends on AI that can perceive, reason, and act in complex real‑world environments. It also requires development workflows that support rapid training, testing, and improvement at scale.

NVIDIA is releasing NVIDIA Alpamayo, a new family of open models, simulation tools, and large datasets to advance reasoning‑based autonomous‑vehicle development. It includes:

• Alpamayo 1 – the first open, large‑scale reasoning VLA model for autonomous vehicles. It enables cars to understand their surroundings and explain their actions.
• AlpaSim – an open‑source simulation framework for closed‑loop training and evaluation of reasoning‑based AV models across diverse environments and edge cases.

In addition, NVIDIA is releasing the Physical AI Open Datasets, which contain > 1,700 hours of driving data collected from the widest range of geographies and conditions. The datasets cover rare and complex real‑world edge cases essential for advancing reasoning architectures.

NVIDIA Clara for Healthcare & Life Sciences

To lower costs and deliver treatments faster, NVIDIA is launching new Clara AI models that bridge the gap between digital discovery and real‑world medicine.

These models help researchers design treatments that are safer, more effective, and easier to produce:

• La‑Proteina – Enables the design of large, atom‑level‑precise proteins for research and drug‑candidate development, giving scientists new tools to study diseases previously considered untreatable.
• ReaSyn v2 – Ensures AI‑designed drugs are practical to synthesize by incorporating a manufacturing blueprint into the discovery process.
• KERMT – Provides high‑accuracy, computational safety testing early in development by predicting how a potential drug will interact with the human body.
• RNAPro – Unlocks the potential of personalized medicine by predicting the complex 3D shapes of RNA molecules.

In addition, NVIDIA has released a dataset of 455 000 synthetic protein structures that helps AI researchers build more accurate models. See the paper on arXiv.

Get Started with NVIDIA Open Models and Technologies

NVIDIA’s open models, data, and frameworks are now available on:

• GitHub
• Hugging Face
• A variety of cloud, inference, and AI‑infrastructure platforms
• build.nvidia.com

This gives developers flexible access to supporting resources.

Many of these models are also offered as NVIDIA NIM microservices for secure, scalable deployment on any NVIDIA‑accelerated infrastructure—from the edge to the cloud.

Learn more by watching NVIDIA Live at CES.