Published: 2026-06-07

NVIDIA & Kilo Code Live Webinar: GPU-Accelerated AI Development

Chapters / key moments (click to jump — plays here on the page)

Live webinar co-hosted by NVIDIA and Kilo Code covering AI-powered development workflows with GPU acceleration. Kilo Code supports 500+ models across 60+ providers — including NVIDIA NIM endpoints — through its native BYOK gateway with zero markup. For local GPU setup, see NemoClaw local GPU guide.

Watch on YouTube → · ← Back to Kilo Code News

What you can actually set up from this

Extracted from the video's own transcript — the specifics the original summary left out.

Reproducible steps

Use a system of models
NVIDIA's suggestion: Nemotron 3 Ultra as the thinker/planner, Nemotron 3 Nano as the fast executor for small jobs (unit tests, grepping files, parsing context). Kilo lets you route by task.
Split interactive and offline work
Fast models suit live pair-programming in the terminal or IDE; slower, heavier runs suit tasks you set and review later.

Gotchas

Nemotron 3 Ultra, as described by NVIDIA's Chris Alexiuk (June 2026): ~550B total parameters with ~55B active (mixture of experts), a hybrid Mamba-2 + attention design for a 1M-token context, NVFP4 pre-training, and fewer tokens per turn than earlier Nemotrons — which can mean more loop iterations on very complex reasoning.
Trained across many agent harnesses (including Kilo, OpenHands, OpenClaw and Hermes) deliberately, rather than tuned to one; Kilo reports it led open-weight models on PinchBench, an OpenClaw agent-skills benchmark.
Automatic routing between models is still unsolved — complexity-, intent- and domain-based routers are, in his words, band-aids, because new models arrive weekly. NVIDIA has a router blueprint.