Ask the Experts: Nemotron 3 Ultra | Nemotron Labs

Nemotron 3 Ultra is here - and in this Ask the Experts livestream, you'll get direct access to the NVIDIA AI researchers behind it. Ultra is NVIDIA's frontier-intelligence open model for long-running autonomous agent workloads: 5x faster inference, up to 30% lower cost, and fully open weights, training datasets, and fine-tuning recipes included. We'll cover the architecture behind Ultra's performance, how it fits into a system-of-models pattern, and how NemoClaw and OpenShell give developers runtime-level controls for always-on agents. What you'll learn: - How Ultra's hybrid Mamba-Transformer backbone handles long-context agent workflows efficiently - How MOPD training delivers consistent SWEBench performance across Hermes, OpenClaw, OpenCode, and LangChain - How to fine-tune Ultra using the open LoRA, SFT, and RL recipes on your own domain data -How NemoClaw and OpenShell provide runtime-level controls for long-running autonomous agents Have questions about Ultra's architecture, training, or deployment? Drop them live — the NVIDIA research team will answer them in real time.