Field noteOct 15, 20262 min read1 source

Nemotron 3.5 Super VL is now free on OpenCode 120B total · 12B active · multimodal NVIDIA's latest open model, built for long-running agents

In one minute

NVIDIA’s release of Nemotron 3.5 Super VL is an attempt to package multimodal understanding and agent-oriented behavior in an open model that organizations can adapt and run on their own infrastructure.

The source describes a 120-billion-parameter model and reports availability through Hugging Face, OpenRouter, ModelScope, and NVIDIA NIM. NVIDIA also says weights, datasets, and training recipes are open. The central thesis is: the model release matters less as a standalone benchmark claim than as another attempt to make an adaptable model fit into a controllable production harness.

The mechanism is a combination of multimodal input handling, tool use and multistep task support, plus deployment and post-training options intended to let organizations adapt the system to their own workflows. The mechanism is: open weights and training materials, combined with deployment on an organization’s infrastructure, give teams a path to adapt a model around their own data and workflows rather than relying only on a hosted endpoint. The release arrives as agent systems increasingly depend on models that can interpret documents and visual material while acting through tools. Why now: the release targets workflows where agents must move between documents, visual evidence, and actions rather than handle text alone.

For builders, the implication is that model choice may be only one part of the decision; integration, evaluation, runtime constraints, and the ability to customize will determine whether the model is useful in production. First implication: teams evaluating agent models should test the complete workflow—including visual inputs, tool calls, and deployment constraints—not infer production suitability from release claims alone. Evidence boundary: These capabilities, comparative performance, and deployment benefits are described by NVIDIA; the supplied material contains no independent evaluation, and it does not confirm the headline’s claim of free access on OpenCode. First watch condition: look for independent evaluations and concrete reports of teams deploying or customizing Super VL in production, with workload and infrastructure details. independent evaluations and verifiable deployment reports will show whether the claimed capabilities translate into useful production workflows. This begins a distinct record about the Super VL release and its positioning; no prior New Runtime record was supplied for this story.

Evidence / sources

  1. 1blogs.nvidia.com/blog/nemotron-3-5-super-vl
All Field Notes

Next reads

Read next

Open the graph
  1. 01topicLong Running AgentsExplore the long-running-agents topic hub.
  2. 02archiveField NotesOpen the latest editorial analysis.
  3. 03source ledgerSource LedgerInspect the public source evidence graph.
Who read this page?Machine requests, hidden until opened

Loading the privacy-safe route aggregate...

Open the JSON contract