What NVIDIA announced
NVIDIA’s October 15 announcement adds Nemotron 3.5 Super VL and Bolt 74B-A9B to its model family. The company says both are available now, while Bolt E4B is planned for October 23. Super VL handles visual and text inputs; these are separate models and deployment targets, not one download that fits every device.
What to check before downloading
The publisher’s Super VL NVFP4 and Bolt 74B-A9B GGUF model cards list the OpenMDW 1.1 license. That is a specific license to review, not a reason to assume Apache 2.0 terms apply. NVIDIA’s performance and efficiency comparisons are vendor results. We have not downloaded, run, or independently benchmarked these releases.
Our take
Editorial analysis: “local” is a location, not a memory requirement. Start with the exact checkpoint, supported runtime, and complete memory needs before comparing a headline score. A model’s active parameter count describes part of its computation; it does not tell you the size of all its stored weights. For a first experiment, a narrower task that runs reliably on existing hardware is more useful than an upgrade bought for an untested promise.