In brief
- Nemotron 3.5 Super VL is an open-weight, 120-billion-parameter model designed for enterprise agents.
- Nemotron 3.5 Bolt 74B-A9B 24GB targets local tasks such as multistep reasoning, tool calling and web browsing.
- Super VL and Bolt 74B-A9B are available; NVIDIA says Bolt E4B will arrive later this month.
- NVIDIA is targeting local use of the Bolt models on AI PCs with RTX Spark chips and DGX Spark workstations.
On Thursday, 15 October, NVIDIA announced new open-weight models from the Nemotron 3.5 family for enterprise agents and local AI use. According to the company, the 120-billion-parameter Nemotron 3.5 Super VL is aimed at enterprise use, while two smaller Bolt models are designed to run locally on computers. Super VL and Bolt 74B-A9B are now available, while Bolt E4B is planned for release toward the end of the month.
Super VL targets enterprise agents
NVIDIA says Super VL can deliver accuracy comparable to Nemotron 3 Ultra on instruction-following and tool-use tasks, despite having fewer parameters. This is the company’s assessment; the source does not include independent test results or comparative measurements.
The model is said to be suitable for areas such as document analysis, financial accounting, scientific experiments, health insurance claims and resilience to security attacks. NVIDIA claims that, after further training for specific agent tasks, Super VL runs faster than models from Anthropic and OpenAI. The company’s speed claim is also contingent on further training for particular tasks.
According to NVIDIA, Skydance, EY, Ramp, Crowdstrike, Perplexity and Salesforce are among the organizations using Super VL. Its open-weight nature is intended to let organizations train it further on their own data and deploy it on their own infrastructure. However, the source does not specify licensing terms or usage fees.
Bolt models are designed to run on computers
The smaller members of the family, the Bolt models, are intended to run AI agents on local systems rather than in the cloud. NVIDIA links this approach to users being able to handle everyday tasks without sending them to the cloud.
Nemotron 3.5 Bolt 74B-A9B 24GB is designed for agent tasks such as multistep reasoning, tool calling and web browsing. NVIDIA says the model aims to balance speed and reasoning. The smaller Bolt E4B is aimed at conversational agents and is due later this month.
The company is targeting the Bolt models for use on new AI PCs with NVIDIA RTX Spark chips and DGX Spark workstations. The article says these devices have begun coming to market through partners such as Microsoft and Dell. Pricing, detailed system requirements and licensing terms for the models have not been announced.
Potential impact on architecture and visualization workflows
For architecture firms and visualization teams, the announcement’s main significance is the option to run agents on local computers or on an organization’s own infrastructure. Tasks such as working with documents, repetitive information work and tool use are among the models’ intended applications; however, the source does not say that ready-made integrations are available for Revit, 3ds Max, Blender or other design software. The models should therefore not be treated as if they can already be connected directly to project workflows.
If considering local use, teams should separately test compatibility with the intended hardware and performance on real workloads. Since the announcement does not provide general system requirements, compatibility with hardware other than RTX Spark or DGX Spark cannot be assumed. It would also be premature to make decisions about enterprise deployment before licensing terms and data-handling options are clear. “Open-weight” does not, by itself, mean free to use.
Availability and claims to consider carefully
NVIDIA says Super VL and Bolt 74B-A9B are available, while Bolt E4B will arrive later this month. The company’s performance comparisons are based on specific agent tasks and further-training conditions; they are not presented as measured results for architectural design, rendering or modeling applications.
A practical next step for teams is to first check licensing and hardware requirements, then test the model on a real document or tool-use scenario. Offices seeking to connect it to design software should wait for integrations to be announced and verify how internal data is handled.
Sources
1 sourceSource texts are not republished; short quotes are marked, everything else is our own summary and commentary.
The ability to run open-weight models within an organization could be a notable option for offices that do not want to send project documents to a cloud-based agent. But the announcement provides no ready-made connections to design software, detailed hardware requirements or licensing fees, so it cannot yet be considered a direct production tool for architecture teams.
For offices in Turkey, the most sensible approach is to first clarify usage rights and hardware compatibility, then run a limited, measurable trial. NVIDIA’s speed claims should not be interpreted as rendering or modeling performance; any benefits need to be tested separately in real workflows.
Frequently asked questions
What is Nemotron 3.5 Super VL?
It is an open-weight, 120-billion-parameter model NVIDIA announced for enterprise agents. The company says it focuses on instruction-following and tool-use tasks.
When will Nemotron 3.5 Bolt models be available?
According to NVIDIA, Bolt 74B-A9B 24GB is available. Bolt E4B is expected later this month.
Which computers will run the Bolt models?
NVIDIA is targeting local use on AI PCs with RTX Spark chips and DGX Spark workstations. The source does not provide compatibility information for other systems.
