NVIDIA Boosts AI Agents, Robotics with New Reasoning Models

NVIDIA has announced significant advancements in its AI model families, Nemotron and Cosmos, designed to power more intelligent AI agents and physical robotics. In an announcement on its blog, the company detailed how these expanded reasoning capabilities are being adopted by major enterprises like CrowdStrike, Uber, and Zoom, aiming to unlock substantial productivity gains and cost savings, with Capgemini projecting up to $450 billion in value by 2028.

The core of this expansion lies in two distinct model families. Nemotron is tailored for enterprise AI agents, providing the "brain" for complex, multi-step tasks within business workflows. The new Nemotron Nano 2 and Llama Nemotron Super 1.5 models are highlighted for their accuracy and efficiency across scientific reasoning, math, coding, and instruction-following. NVIDIA claims these models offer leading efficiency through a hybrid architecture, compact quantized models, and a configurable "thinking budget," resulting in up to 60% lower reasoning costs and faster, deeper responses.

For physical AI applications, NVIDIA introduced Cosmos Reason, a new reasoning vision language model (VLM). This model is purpose-built to help robots and vision AI agents understand and interact with the real world, excelling in concepts like physics, object permanence, and spatial-temporal alignment. It serves as a crucial backbone for robot vision language action (VLA) models, enabling more human-like reasoning for tasks in factories, warehouses, or autonomous vehicles.

Reasoning for the Real World and the Enterprise

The practical applications of these models are already taking shape across diverse industries. Zoom plans to integrate Nemotron reasoning models into its AI Companion to manage multi-step tasks across its communication platforms. Cybersecurity firm CrowdStrike is testing Nemotron to enable its Charlotte AI agents to generate queries for the Falcon platform, while NetApp is exploring its use for searching and analyzing business data. Other enterprise adopters include Amdocs, EY, DataRobot, and Tabnine, all leveraging Nemotron for various agentic AI solutions.

On the physical AI front, Uber is exploring Cosmos Reason to analyze autonomous vehicle behavior, including summarizing visual data and understanding complex scenarios like pedestrians crossing highways. Magna is developing Cosmos Reason for its City Delivery Platform to help autonomous vehicles adapt to new environments. Companies like Centific and VAST are using Cosmos Reason to enhance video intelligence platforms for public safety and urban monitoring, while Ambient.ai is applying it for environmental health and safety in industrial settings, detecting missing personal protection equipment and hazardous conditions.

NVIDIA emphasizes that these models bridge the gap between raw AI intelligence and practical business utility by providing leading libraries and AI blueprints for onboarding, customizing, and governing AI agents at scale. The new models are expected to be available as NVIDIA NIM microservices for secure deployment, with planned availability through major cloud platforms like Amazon Bedrock, Amazon SageMaker AI, Azure AI Foundry, Oracle Data Science Platform, and Google Vertex AI. Cosmos Reason is already accessible via build.nvidia.com, Hugging Face, and GitHub, with the new Nemotron models slated for release soon.