The Nvidia Open AI model portfolio spans the scientific world. NVIDIA
Nvidia, normally thought of as favoring closed software like CUDA, has actually become the leading source for open models across a broad industry swath, from Agentic AI to Autonomous Vehicles and drug discover. The company recently held and analyst session to go over the portfolio of seven unique AI models it has developed and shared specific use cases for each. While not “new news”, I found the overview helpful and hope you do as well. These open models, I believe, help demonstrate Nvidia’s leadership in the development of AI solutions with their partners, which of course drags along a lot of Nvidia hardware. (Note that Nvidia is a client of my firm, Cambrian-AI Research, LLC.)
Open Versus Closed Models
It is not really one versus another, as the foundation model developers like OpenAI and Anthropic fill an important leadership role in AI model research and advancements. Closed models will remain the gold standard, but closed model fill in the domain-specific gaps the closed models leave open, and do so with an advantageous cost model. All these models here are free and open to use. Nvidia does sell the Nvidia AI Enterprise AI license for the tools to use and to run these models.
Open and Closed models each fill a specific need in the AI marketplace. NVIDIA
The Nemotron Family of Open Models for Agentic AI
Nemotron has been around since before agentic AI became a thing, and has served as Nvidia’s primary model for generative AI. Since Agents hit the scene, it has served as the standard for many agentic AI services.
Nemotron was Nvidia’s first open generative AI model and is often used for long running AI agents. NVIDIA
One example use case is Dassault Systèmes Super Agent, which is used in the design and analysis of physical CAD engineering, In this case Dassault uses Nemotron to design car doors, resulting in 28% higher design quality, 25% productivity improvement across a broad set of engineer design tasks and a 35% reduction in tool calls.
Nvidia Cosmos for the World of Physical AI
Cosmos has become the industry’s foremost AI model for vision reasoning, world and action generation. It understands physics, unlike Large Language Models (LLMs) which are built on words and sentences. It helps engineers design custom physical AI models for edge deployment in automobiles, robotics, and factories.
Nvidia Cosmos for Physical AI. NVIDIA
An example here is Siemens which uses Cosmos to run Always-On Factory Agents, which targets a 10-fold speedup for video inference processing with up to 85% lower engineering effort and 40% fewer false calls.
Nvidia Isaac GR00T for General Purpose Humanoid Robotics
Humanoid robots could be the Next Big Thing. At Davos in January 2026, Elon Musk predicted a future with more robots than people, accounting for some 80% of Tesla’s value. Getting there requires a robotic training far beyond today’s early prototypes, and Isaac GR00T is “here to help”. GR00T (those are zero’s, not “o”s) helps build robotics with a high degree of dexterity for actions, with policy-driven actions for embodiments, environments, and tasks.
Nvidia Isaac Groot is a vision-based action model for high-dexterity manipulation. NVIDIA
An example here is Unitree H2, which is an open humanoid reference design that offers 75 degrees of freedom across the body and hands and has attracted a rich ecosystem of partners around the world.
Nvidia Alpamayo for Autonomous Vehicle Development
Alpamayo has rapidly become the leading model to help companies speed development of the brains behind autonomous vehicles. Auto companies can no longer start from scratch, and Alpamaoy has already garnered support from BYD, GM, Hundai, JLR, Li Auto, Lucid and Mercedes Benz. Alpamayo provides the simulation environment for developing and testing road-ready smart vehicles with the security needed for automated vehicles.
Nvidia Alpamayo is the company’s open model for autonomous vehicle development. NVIDIA
Nvidia’s poster-child for Alpamayo is Mercedes Benz, which has rolled out Drive Pilot 3, enabling SAE Level 3 and 2+ using custom software and Alpamayo. Mercedes is now working on Robotaxi designs to enable full Level-4 autonomy with Nvidia DRIVE, automating over 95% of urban driving decisions using Alpamayo.
Nvidia BioNeMo for Life Sciences
Life Sciences spans diverse markets like drug discovery from proteins, RNA therapy designs, drug molecules, and chemical recipes, each with its own AI model to speed development. Here, Nvidia has teamed with over eight companies that use BioNeMo to accelerate drug design and manufacturing innovations.
Nvidia BioNeMo advances the use of AI for healthcare innovation. NVIDIA
The example Nvidia shared here was Lila Scientific Superintelligence, which has realized a rather shocking 98% success rate for generating experimental designs with Nemotron 3.5 Super compared to Nemotron 3 Super in generating lab-ready experiments and over 950K mRNA sequences designed, built, and tested in living cells.
Earth-2 delivers the first Open AI Weather Stack
I don’t know if you’ve noticed, but weather foreacsting has improved considerably over the last few years. Nvidia’s Earth-2 delivers models and tools for fine-scale high-accuracy weather forecasting, and has become useful in global medium-range forecasting and now casting. Nvidia claims it can produce 10X faster local forecasts at 0-% lower cost. The Israel Meteorological Service uses Earth-2 models and tools to produce high-resolution forecasts and reports a 90% reduction in compute time compared with a traditional weather model running on a CPU cluster.
Nvidia Earth-2 is for AI physics. NVIDIA
Nvidia Ising Advances Quantum Computing
Two of quantum computing’s biggest challenges remain in improving error detection and correction and processor calibration. It turns out, AI can help here as well, and Nvidia has built an open model family already in use. Nvidia is teaming with a large number of quantum computing system companies to help, including IQM, Q-Ctrl, and Infleqtion. The Ising open models help automate calibration and decoding of quantum hardware, regardless of the qubit modality being used (superconducting, neutral atoms, trapped Ions, etc.).
Nvidia Ising helps advance Quantum Computing. NVIDIA
Onbe proofpoint here quantum processor builder Aegiq, which is accelerating processor calibration time by three-fold with Ising, while improving decode by 2.5X.
The Bigger Message
Nvidia hardware of all sorts, AI and HPC software, Open AI models and now financing has helped Nvidia blunt some of the impact of a world increasingly full of GPU alternatives. Whether a Google TPU or an AMD GPU is as fast or faster than an Nvidia GPU is no longer the deciding purchase factor. Customers are buying whatever can get them up and running on new AI models to solve business and scientific challenges. Nvidia’s Open Source portfolio is yet another reason many customers and workloads will remain on Nvidia GPUs for the foreseeable future.