NVIDIA AI

One model is no longer enough for complex AI workflows. The teams getting the most out of agentic AI are building systems of models – routing each step to the right intelligence based on task complexity, cost, latency, and quality. The question isn't which model to pick. It's how to route intelligently across all of them. This session covers how NeMo Switchyard, an open source model routing library, brings system-of-models routing to real agent workflows. We'll start with why routing intelligently across any combination of open and closed models gives your stack more flexibility and control, then get hands-on: Cognition will show how they've integrated NeMo Switchyard to route coding agent workflows — from planning and coding to testing, debugging, and review — automatically selecting the right model at each step. What you'll learn: Why a system of models outperforms a single-model stack for complex agent workflows How NeMo Switchyard routes each agent step to the right model, helping agents complete work faster while balancing accuracy, control, and efficiency How Switchyard routes across any combination of open and closed models, and why that flexibility gives your stack more control How Cognition integrated Switchyard into a coding agent platform Building with a system of models and thinking about how to route across your stack? Bring your questions – Cognition and the NVIDIA team will answer them live.

As AI agents take on more complex work, where should security live? The harness guides what an agent tries to do. Infrastructure controls what it can actually do. Our AI safety and security teams share how they’re thinking about where security fits across the agent stack, and the controls needed as agents become more capable. Read the deep dive: https://nvda.ws/4zsLztl

Our general-purpose coding agent just scored 100% on the ARC-AGI-3 interactive reasoning benchmark. NVIDIA AVO completed all 183 levels across all 25 public environments, figuring out what to do with no instructions, explicit rules, or stated goals. AVO continuously inspects, plans, implements, and evaluates, using memory, tools, and execution feedback to build on what it learns along the way. This allows the system to sustain progress across long-running tasks rather than starting over with each model context. Read about AVO and how we built it for long-horizon autonomous agents: https://nvda.ws/4qqFLMG

In this Cosmos Labs livestream, NVIDIA experts, Aigen, and Linker Vision will demonstrate how domain-specific data can specialize Cosmos for robot policies, vision-language models, and faster world generation—and how to move those models from training through evaluation and deployment. Explore the full post-training journey through real-world examples from Aigen and Linker Vision. Aigen will show how it adapts Cosmos for autonomous agricultural robotics, while Linker Vision will demonstrate how specialized VLMs improve video reasoning in complex physical environments. NVIDIA experts will connect these examples to practical workflows, including Cosmos3-DROID policy post-training and Cosmos 3 Super step distillation for faster synthetic data and world generation. What you’ll learn: - How to prepare domain-specific data and post-train Cosmos 3 - How to improve robot policy learning with Cosmos3-DROID - How Linker Vision specializes VLMs for real-world video reasoning - How step distillation accelerates synthetic data and world generation - How Aigen applies an end-to-end Cosmos workflow to agricultural robotics - Best practices for evaluating and deploying specialized Cosmos models Have questions about post-training, optimization, evaluation, or deployment? Join live and ask the NVIDIA, Aigen and Linker Vision teams. 📖 Read the technical walkthrough for Cosmos 3 Edge post-training → XX 🎥 Watch the tutorial → XX 🔴 Join the Cosmos Labs livestream and ask questions live → XX 🤗 Read the Hugging Face blog → XX 📥 Download Cosmos 3 → https://lnkd.in/gE_uy_jT 🐙 Customize Models with Cosmos 3 → https://lnkd.in/gRY3QEvU

NVIDIA cuOpt is the fastest open source solver on Hans Mittelmann benchmarks across three optimization problem classes. Check out NVIDIA cuOpt on GitHub and let us know what you build with it: https://lnkd.in/gNSNkH-M

We benchmarked 300+ NVIDIA verified skills to see how much they actually help agents on real tasks. Same task, same model, same setup. The only difference was whether the agent had the skill. Across the benchmarks, skills improved correctness by 41 points, effectiveness by 39, and efficiency by 35. SkillEvaluator is open source if you want to test your own skills before you ship them: https://nvda.ws/46c2R0c

Are you building something exciting with open models? I’ve partnered with NVIDIA to give AI builders a shot at a Golden Ticket to #NVIDIAGTC Berlin. 🎫 If you’re building with open models, show us what you’re working on, you could be joining us in Berlin. Your Golden Ticket includes: → Free conference pass → VIP seating for Jensen Huang’s keynote → Exclusive NVIDIA merch → Access to special events How to enter: Share your project below. 📅 Submissions: Aug 18 – Sep 10, 2026 🏆 Winner announced by September 14 👉 Contest details: https://nvda.ws/4hP8W9S Building something worth seeing? "Comment" your project below.👇 Or tag an AI builder who should be in Berlin.

What are you building with open models? Show us what you’re working on and you could be headed to #NVIDIAGTC Berlin. 🎫 Your Golden Ticket includes: → Free conference pass → VIP seating for Jensen’s keynote → Exclusive NVIDIA merch → Access to special events Submissions are open Aug 18 – Sep 10, 2026. Details and how to enter: https://nvda.ws/4ieFwCm

What goes into building an open model for long-running agents? In this Ask the Experts session, you'll get direct access to the NVIDIA AI researchers building and training the latest Nemotron open models. Join us to hear what's driving their design decisions: how these models are trained to handle specialized, high-volume agentic tasks, why Nemotron open models are built to run locally — on hardware like DGX Spark — without sacrificing accuracy or speed, and what it means to build an open model that fits into a system where different tasks call for different models. We'll cover how the research team approaches training open models purpose-built for agentic work: fast, efficient, and designed to be customized for your domain. The conversation will dig into the architecture choices, training techniques, and design principles behind models built to run anywhere — from DGX Spark to the data center — and to slot cleanly alongside other models in a multi-model agent system. What you'll learn: How NVIDIA AI researchers train open models for long-running, autonomous agents What makes the Nemotron family customizable for domain- and task-specific workflows How Nemotron open models are optimized to run locally, from DGX Spark to your own infrastructure How Nemotron open models are built to work in systems where different steps call for different models Have questions about open model training, agentic design choices, or running Nemotron locally? Drop them live — our researchers will answer them directly.

We just released TensorRT Model Connect in Public Preview. You can take a supported Hugging Face model to end-to-end TensorRT inference in just two commands. No intermediate ONNX export, and the resulting bundle can run through native C++ APIs. We also built the entire project with OpenAI Developers Codex agents, with humans directing and reviewing the work. That includes model implementations, performance tuning, tests, integrations, and docs. It’s open source, so go try it out, dig into the implementations, or contribute support for a new model: https://lnkd.in/dAHCMX8S