Key Insights

  • Nvidia Nemotron 3.5 adds open models for data centers, 24GB local systems and compact edge devices.
  • Super VL has 120 billion parameters and claims five-times faster throughput than comparable open models.
  • Bolt 74B-A9B is available now, while the 4GB Bolt E4B model arrives October 23.

Nvidia Nemotron 3.5 expanded the company’s open-model lineup across enterprise servers, local workstations and edge devices. Nvidia released Super VL and Bolt 74B-A9B on October 15, while Bolt E4B is scheduled for October 23.

The release targets organizations that want customizable agents without sending every workload to a third-party cloud. Nvidia said the models include open weights, training datasets and development recipes.

Nvidia Nemotron 3.5 targets enterprise agents

Nemotron 3.5 Super VL is a 120-billion-parameter vision-language model built for tool use, multistep workflows and document analysis. Nvidia said it delivers five-times faster throughput than the next-best model in its class.

The company also said the model runs at roughly one-fifth the parameter count of its largest reasoning model. Those are vendor benchmarks, so customers should test performance against their own workloads and hardware.

Bolt moves agent workloads to local hardware

Bolt 74B-A9B uses native 2-bit quantization to fit within 24GB of memory. Nvidia designed it for RTX and DGX Spark systems running research, writing and software-development agents locally.

Bolt E4B targets 4GB deployments for conversational and physical agents. Local processing can reduce cloud dependence, but performance still depends on memory, software support and task complexity.

What investors and developers should watch

Developers should compare model quality, inference speed and memory use before deployment. They should also examine update support, security controls and the cost of post-training specialized agents.

The release extends Nvidia’s push across the AI stack. The company recently introduced tools for AI-defined semiconductor factories, while its GPUs supported faster LSEG market-data analysis.

Security remains important as agents gain access to enterprise systems. Radware’s recent MCP server protection launch shows how vendors are building controls around these workflows. The official Nvidia announcement lists the models, availability dates and performance claims.

Sharron Kendi is a seasoned crypto market analyst and writer with over three years of experience decoding blockchain trends, price movements, and market dynamics. She holds a Bachelor’s Degree in Commerce (Finance) from Kenyatta University, blending a solid academic foundation with a sharp eye for technical analysis and a deep understanding of on-chain data. Her work delivers clear, data-driven insights that empower investors to navigate the fast-evolving digital asset space with confidence.