Hajra Naz
Nvidia introduced new open-weight AI models on Thursday specifically built for always-on autonomous agents and edge computing.
The primary release is the Nemotron 3.5 Super VL. This large 120-billion-parameter model targets enterprise-grade autonomous workflows. Although it uses fewer parameters than the earlier Nemotron 3 Ultra model, Nvidia claims it delivers comparable accuracy for instruction-following and tool-use tasks.
Bringing Local AI Processing to Spark-Powered Workstations
At the other end of its product lineup, Nvidia unveiled two lightweight Bolt models. These smaller architectures are engineered to run directly on local hardware. They target next-generation AI PCs powered by Nvidia RTX Spark chips and workstations equipped with DGX Spark hardware.
Hardware manufacturers are actively shipping these consumer and developer systems to market. Key retail and manufacturing partners include Microsoft and Dell. The devices target everyday consumers, software engineers, and corporate knowledge workers.
Nvidia Vice President of Open Models and Software Kari Briski outlined the strategy in an official announcement. She stated that any organization deploying agents on Nvidia hardware now has access to an optimized open model tailored to their specific deployment tier, balancing operational capability with ease of deployment.
Read More: Amazon makes its boldest move against Nvidia with new TorchNeuron dev tool
Performance Benchmarks and High-Profile Corporate Partners
Nvidia claims that the Super VL model completes complex workflows faster than proprietary alternatives from Anthropic and OpenAI, including Opus 5, GPT 6.1 Sol, Luna, and Astra. These performance gains appear after fine-tuning the model on specialized tasks. Key use cases include document processing, financial accounting, scientific research, health insurance claims, and cyber defense resilience.
Tech companies routinely highlight bold performance benchmarks during major product launches. However, Nvidia supported its claims by naming major corporate clients that are already using Super VL. Early enterprise adopters include Skydance, EY, Ramp, Crowdstrike, Perplexity, and Salesforce.
This release builds on Nvidia’s expanding portfolio of open-weight offerings. The GPU manufacturer continues to compete directly with proprietary AI labs, many of which remain its key customers and investment partners. Nvidia maintains a broad family of Nemotron models designed for multimodal analysis, logical reasoning, software coding, mathematics, and business automation.
Read More: Microsoft Launches Nvidia-Powered AI PCs With a Revamped Windows 11
Addressing Cloud Competition and Safeguarding Data Privacy
Rapidly growing AI research labs are actively seeking hardware options beyond Nvidia to meet their massive compute demands. Many developers are shifting workloads toward alternative platforms like AWS Trainium, Cerebras, or proprietary internal silicon.
In response, Nvidia is focusing heavily on corporate clients that prefer running open models within their own data centers. Briski emphasized that unlike closed proprietary models, businesses can fine-tune Super VL on proprietary internal data, host it on local infrastructure, and retain full intellectual property rights.
Simultaneously, Nvidia is targeting the emerging market for local consumer AI assistants. While tech startups launch new digital assistants daily, privacy-conscious consumers often hesitate to share sensitive data with cloud platforms. Many users remain reluctant to upload family information, medical records, or personal financial details to remote servers.
Nvidia aims to open new revenue streams by marketing specialized AI PCs powered by its silicon. These local systems allow users to execute daily tasks and delegate project workflows without transmitting sensitive personal data to external cloud servers.
Read More: Nvidia Launches New Platform to Control Rogue AI Agents
Specifications and Global Availability of the Nemotron Lineup
The company detailed specific configurations for its new lightweight models:
Nemotron 3.5 Bolt 74B-A9B (24GB): A 74-billion-parameter architecture optimized for local agent tasks. It balances processing speed with complex reasoning, supporting multi-step problem solving, external tool integration, and web browsing.
Nemotron Bolt E4B: The smallest variant in the lineup, optimized specifically for fast, conversational AI interactions.
Nvidia confirmed that both the Super VL and Bolt 74B-A9B models are available immediately for developer deployment. The smaller Bolt E4B model is scheduled to roll out later this month.
