Introducing NVIDIA Nemotron 3.5 Super VL

Summary
This is an official product announcement video from NVIDIA Developer introducing NVIDIA Nemotron 3.5 Super VL. Presented without voiceover using kinetic text overlays and UI mockups set to background music, the video demonstrates the open multimodal model executing agentic tasks, including hazard monitoring, sketch-to-simulation coding, and chart cost analysis.

What is shown

• [00:01 - 00:10]: Title cards announcing “NVIDIA Nemotron 3.5 Super VL” as an “Open Multimodal Model Built for Specialized Agents”, highlighting post-training, visual understanding, and tool calling.
• [00:11 - 00:16]: A terminal/agent interface (“Hermes Agent”) executing a morning briefing and inventory/safety monitoring workflow (“Spot safety and inventory issues.”).
• [00:17 - 00:36]: “Spot Hazards in Videos”: In response to a prompt (“Please monitor my store and alert me of anything that might require my action”), an isometric store simulation detects a spill in the produce aisle, assesses its severity, and pushes a recommended response plan to Telegram (“Get a recommended next step.”).
• [00:37 - 00:54]: “Turn Sketches Into Interactive Apps”: The model takes physics sketch diagrams (orbital ballet, wave interference, emergent flock, gravity playground, pendulum choir, flow field), tests code, recommends fixes, and renders interactive browser simulations (“Draw an idea. Watch it move.”).
• [00:55 - 01:11]: “Analyze Charts to Explain Costs”: Given a question about rising token bills, the system interprets spend charts and token type distributions, isolating that AI spend rose 30% while token pricing decreased by about half (“Turn charts into clear analysis.”).
• [01:12 - 01:20]: Closing tagline (“Specialized agents. Built for efficiency.”) and NVIDIA logo.

Claims & numbers

• NVIDIA presents Nemotron 3.5 Super VL as an “Open Multimodal Model” specialized in tool calling, post-training, and visual understanding [00:03 - 00:10].
• In the store monitoring sequence, the system identifies a medium-severity wet floor spill and calculates a response team readiness time of approximately 5 minutes [00:33 - 00:36].
• In the cost chart demonstration, the model extracts metrics showing AI spend increased 30% (from $94.4k to $122.7k) while the average price per 1M tokens dropped ~49.6% (from $2.33 to $1.18) [00:58 - 01:07].

Notable quotes

• [00:03]: “Open Multimodal Model Built for Specialized Agents”
• [00:41]: “Draw an idea. Watch it move.”
• [01:13]: “Specialized agents. Built for efficiency.”

Assessment
This is an official promotional product teaser presenting stylized UI mockups, simulations, and polished workflows rather than an unedited live bench test. While it effectively illustrates the intended application domains of the vision-language model, exact performance benchmarks, model weights, and hardware parameters are not displayed within the video itself.

Described by gemini-3.8-flash on 2026-10-15 from the video’s audio and frames.