Nvidia Ships Three Open Nemotron Models at Once — 120B Vision-Language Down to 4GB Edge — Covering Every Deployment Tier the Week of GTC Berlin

Nvidia Ships Three Open Nemotron Models at Once — 120B Vision-Language Down to 4GB Edge — Covering Every Deployment Tier the Week of GTC Berlin

New subject (announced Oct 15, 2026). Nvidia expanded its open model family with three releases in one announcement — "NVIDIA Expands Open Model Families to Power the Next Wave of Agentic, Physical and Healthcare AI" (Nvidia IR) — spanning data center to 4GB: Nemotron 3.5 Super VL (120B parameters, open vision-language, aimed at healthcare and physical AI), Nemotron 3.5 Bolt 74B-A9B (24GB footprint, edge model "built for general agentic tasks, including navigating web browsers and operating applications"), and Nemotron Bolt E4B (4GB, "physical and conversational agents" for robotics/embedded).

The notable fact is what's missing: "Nvidia has not published benchmark scores alongside the announcement, so direct performance claims remain unconfirmed pending independent testing," and no pricing or licensing terms appear in the available reporting. This is a breadth play, not a leaderboard play — "instead of one model trying to serve every use case, Nvidia is now shipping purpose-built variants across the entire size spectrum simultaneously."

The strategic logic is chips: "Every model that runs well on Nvidia hardware... reinforces demand for Nvidia GPUs. An open model is also a hedge against closed competitors locking developers into their own ecosystems.¹" The open-weights push sits on top of the reported $13B Hugging Face acquisition (September) and follows Nemotron 3 Ultra (550B) and Nemotron 3.5 Lightning (30B).

Timing: the drop landed inside the same 72-hour window as Microsoft's MAI-Base-2 disclosure (Microsoft's MAI-Base-2: A 2.8T-Parameter Statement of Independence From OpenAI — With Every Headline Claim Self-Reported and Nothing Shipped), Meta's Muse Spark Open (Meta Ships Muse Spark Open: 816B Open Weights Claim the US Open-Weight Crown — While the 17-Point Gap to Meta's Own Closed Model Is the Business Model), and MBZUAI's Jais 3 — and days before GTC Berlin (Oct 20–22, Jensen Huang keynote Oct 21), where Nvidia's own blog previews the open-model/agent story ("Anyone Can Build an Agent"). BenchLM's tracker counted 22 confirmed model launches in October by Oct 15 (via Tech-Insider's Liquid AI coverage) — the densest open-weight month on record.

What it means: Nvidia is the only player shipping across the whole deployment curve — cloud, desktop, embedded — so every open-weight deployment, including rivals', still tends to run on Nvidia silicon. The check: independent benchmarks for Super VL and the actual license terms, which will determine whether this is a developer magnet or shelfware.