According to Capgemini, AI agents are poised to provide up to $450 billion by 2028 from revenue growth and cost reductions. Developers building these agents are turning to better performance inference models to improve AI agent platforms and physical AI systems.
On Siggraph, Nvidia today announced the expansion of two model families with the inference capabilities of Nvidia Nemotron and Nvidia Cosmos.
Crowdstrike, Uber, Magna, NetApp and Zoom are some of the companies that use these model families.
The New Nvidia Nemotron Nano 2 and Llama Nemotron Super 1.5 models offer the highest accuracy in the Scientific Inference, Mathematics, Coding, Tool Call, Teaching Publication, and Chat Size categories. These new models empower AI agents to think more deeply and work more efficiently. Explore a wider range of options, speed up your research and deliver smarter results within set time limits.
Think of the model as the brain of an AI agent. Provides core intelligence. But to make that brain useful for your business, you must be embedded in agents who understand and operate specific workflows in addition to industrial and business terminology. Nvidia helps businesses with the major libraries and AI Blueprints to onboard, customize and manage large AI agents to fill that gap.
COSMOS Reason is a new inference vision language model (VLM) for physical AI applications that excels at understanding how the real world works, using structured inference to understand concepts such as physics, object persistence, space-time alignment, and more.
Cosmos Reason aims to serve as the inference backbone of robotics and language action (VLA) models, or as critique and caption training data for robotics and autonomous vehicles, and, like factories and cities, runs-time visual AI agents are equipped with spatial understanding and inference of physical operations.
Nemotron: Agent Enterprise AI is the most accurate and efficient
To develop AI agents and tackle complex multi-step tasks, models that can provide powerful inference accuracy with efficient token generation, enable intelligent and autonomous decision-making at scale.
Nvidia Nemotron is a family of highly open inference models using key models, NVIDIA curated open datasets, and advanced AI technologies to provide an accurate and efficient starting point for AI agents.
The latest Nemotron models provide lead efficiency in three ways: a new hybrid model architecture, a compact quantum model, and a configurable thinking budget that allows developers to control token generation. This combination allows the model to reason deeper and respond faster, without the need for more time or computing power. This means better results at a lower cost.
The Nemotron Nano 2 offers six times higher token generation compared to other major models of its size.
Llama Nemotron Super 1.5 achieves key performance and best inference accuracy in the class, allowing AI agents to enhance better inference, make smarter decisions, and handle complex tasks independently. It is now available in NVFP4 or 4-bit floating point. This provides six times higher throughput on the NVIDIA B200 GPU compared to the NVIDIA H100 GPU.
The chart above shows that Nemotron models provide the highest inference accuracy with the same time frame and the same calculation budget, providing the highest accuracy per dollar.
In addition to two new Nemotron models, Nvidia is also unveiling its first open VLM training dataset (Llama Nemotron VLM Dataset V1) using 3 million optical character recognition, visual QA and caption data on the previously released Llama 3.1 Nemotron Nano VL 8B model.
In addition to the accuracy of the inference model, agents rely on searched generations to obtain the latest and most relevant information from connected data across a variety of sources to make informed decisions. The recently released Llama 3.2 Nemo Retriever Embedding Model is located on top of three Visual Document Retrieval Leaderboards: Vidore V1, Vidore V2 and MTEB VisualDocumentRetrieval, to improve the accuracy of your agent system.
Using these inference and information search models, the deep search agent built using the AI-Q NVIDIA Blueprint is currently number one open and portable agent for DeepResearch Bench.
Nvidia Nemo and Nvidia Nim Microservices support the entire AI agent lifecycle, from development and deployment to monitoring and optimization of agent systems.
COSMOS Reason: Physical AI Breakthrough
VLMS marked breakthroughs in computer vision and robotics, allowing machines to identify objects and patterns. However, non-adaptive VLMs do not have the ability to understand and interact with the real world. This means that it cannot handle ambiguity, new experiences, or solve complex multi-stage tasks.
Nvidia Cosmos Reason is a new open, customizable 700 million parameter inference VLM for physical AI and robotics. For Cosmos reasons, robots and vision AI agents can understand and act in the physical world, using prior knowledge, understanding of physics and common sense, for human-like reasons.
COSMOS Reash enables advanced capabilities across robotics and physical AI applications such as data critique and caption, robot decision making, and training data such as video analytics AI agents.
It helps to automate curation and annotation of large and diverse training data sets and accelerate the development of high-precision AI models. It also serves as a sophisticated inference engine for robot planning, allowing you to analyze complex instructions to practical steps in VLA models, even in new environments.
It also powers video analytics AI agents built on the NVIDIA Blueprint for video search and summary (VSS) enabled by the NVIDIA Metropolis platform, gathering valuable insights from a vast amount of stored or live video data. These visually perceptive and interactive AI agents can help streamline operations at factories, warehouses, retail stores, airports, transport intersections, and more by detecting anomalies.
The Nvidia robot research team uses Cosmos reasons for data filtration and curation, using the “System 2” inference VLM behind VLA models such as the next version of the NVIDIA ISAAC GR00T NX.
Currently offering services: Nvidia inference model for AI agents and robots
A diverse range of companies and consulting leaders employ Nvidia's latest inference model. Leaders across cybersecurity work with Nemotoron to build enterprise AI agents, from telecommunications.
Zoom plans to leverage the Nemotron Reasoning model using Zoom AI companions to make decisions and manage multi-stage tasks that take actions to users in Zoom Meetings, Zoom Chat and Zoom Documents.
CrowdStrike is testing the Nemotron model so that the Charlotte AI agent can write queries to the Crowdstrike Falcon platform.
AMDOCS uses the Nvidia Nemotron model in its AMAIZ suite to drive AI agents to handle complex, multi-stage automated spanning care, sales, networking and customer support.
EY adopts NemotronNano 2 for its high throughput and supports agent AI in large organizations for tax, risk management and financial use cases.
NetApp is currently testing the Nemotron Reasoning model to enable AI agents to search and analyze business data
Datarobot works with the Nemotron model for its Agent Workforce Platform for end-to-end Agent Lifecycle Management.
Tabnine works with the Nemotron model to propose and automate coding tasks on behalf of developers.
Automation Anywhere, Crewai and Dataiku are among the additional agent AI software developers who integrate the Nemotron model into the platform.
Large companies across transportation, safety and AI intelligence use COSMOS reasons to improve autonomous driving, video analytics, and road and workplace safety.
Uber is exploring the universe's reasons to analyze the behavior of autonomous vehicles. Additionally, Uber is the reason for post-training Cosmos to summarise visual data, analyze scenarios such as pedestrians roaming the highway, perform quality analysis, and analyze scenarios such as notifying autonomous driving behavior.
The reason for Cosmos is that it also functions as the brain of self-driving cars. The robot interprets the environment and is given complex commands, so it breaks them down into tasks and runs them using common sense even in unfamiliar environments.
Centrific is testing Cosmos' reasons to enhance its AI-powered video intelligence platform. VLM allows the platform to process complex video data into actionable insights, reduce false positives and improve decision-making efficiency.
Vast uses Nvidia Cosmos reasons to advance real-time urban intelligence to process large video streams using AI operating systems. VSS Blueprints allow giant agents to build agents that can identify incidents, trigger responses, and turn video streams and metadata into actionable and aggressive public safety tools.
Ambient.ai works with Cosmos Reason's temporary physical perception inference to enable automatic detection of missing personal protective equipment and monitoring of hazardous conditions, helping to enhance the health and safety of the environment across construction, manufacturing, logistics and other industrial environments.
Magna is being developed for Cosmos reasons as part of the urban distribution platform. It is a completely autonomous, low-cost solution for immediate delivery, helping to adapt more quickly to new cities. This model adds a global understanding to the vehicle's long-term trajectory plan.
These models are expected to be available as NVIDIA NIM microservices for secure and reliable deployments in accelerated infrastructure to make the most of privacy and control. They will be available immediately through Amazon Bedrock and Amazon Sagemaker AI for Nemotron models, as well as Azure Ai Foundry, Oracle Data Science Platform, and Google Vertex AI.
try Why cosmos burnion.nvidia.com Or download it Hugging my face or github.
Nemotron Nano 2 and Llama Nemotron Super 1.5 (NVFP4) will be available for immediate download. Meanwhile, please check the details of the Nemotron model and download the previous version.
Please download Llama Nemotron VLM Dataset V1 From hugging her face.
Watch Siggraph's Nvidia Research Special Address and learn more about how graphics and simulation innovations can drive industrial digitization by joining Nvidia at the conference, which runs through Thursday, August 14th.
look Let me know Regarding software product information.
