Nvidia releases open-weight Nemotron 3.5 Super VL (120B) for enterprise agents and two Nemotron 3.5 Bolt models for local AI PCs

Nvidia released three open-weight models in its Nemotron 3.5 family. Super VL (120B parameters) targets enterprise agents. Bolt 74B-A9B (24GB) and Bolt E4B target local use on RTX Spark AI PCs and DGX Spark workstations. Super VL and Bolt 74B-A9B are available now, and Bolt E4B is due later this month. Nvidia names Skydance, EY, Ramp, CrowdStrike, Perplexity and Salesforce as Super VL users, but the coverage gives no primary link and no independent test results.

Enterprises that want agents on their own infrastructure now have another open-weight option from a vendor that also sells the hardware to run it. Teams already using Nemotron models can compare the new versions against their current ones. The Bolt models matter mainly to people with Nvidia's new AI PC and workstation hardware, because they offer a way to keep agent tasks off the cloud. The release extends an existing model family rather than opening a new category, so the impact is moderate. Whether it changes buying decisions depends on independent benchmarks and real deployments.

The claim that Super VL finishes tasks faster than Anthropic and OpenAI models "once post-trained" is Nvidia's own and is conditional, so it is not a like-for-like comparison. The claim that Super VL matches Nemotron 3 Ultra accuracy is also unverified. The customer list says who is using or testing the model, not how much, or whether it is in production. The "no trade-off between capability and deployability" line is marketing language.

1. 01Independent benchmarks and third-party evaluations of Super VL against Nemotron 3 Ultra and comparable closed and open models on instruction-following and tool use, including the speed claim with the post-training setup stated.
2. 02Whether Bolt E4B ships by the end of October as promised, and what real-world speed and memory figures reviewers report for Bolt 74B-A9B on RTX Spark and DGX Spark hardware.
3. 03Evidence of production use from the named customers (Salesforce, CrowdStrike, EY and others), such as case studies or product announcements, and download and adoption numbers on model hubs.