NVIDIA used an October 15 release to extend its Nemotron open-model family in two directions: a large vision-language model for agentic workloads and smaller Bolt models intended for local deployment.
The release names Nemotron 3.5 Super VL as a 120-billion-parameter model. NVIDIA describes it as an open model built for visual reasoning and tool-using agents, but its capability and efficiency comparisons in the announcement are the company’s own claims.
The Bolt line targets smaller hardware footprints. NVIDIA says Bolt 74B-A9B is available now and can run in 24GB of memory; the company says the E4B variant, intended for 4GB systems, is expected October 23. Availability dates and deployment claims are reported as announced, not independently tested here.
The strategic shift is toward making capable agent systems deployable beyond data-center GPUs, including workstation and edge environments. Whether developers see the claimed efficiency in practice will depend on task, runtime, quantization and hardware; independent benchmarks were not located in this review.