Three sizes point to different deployments
NVIDIA’s October 15 announcement introduces Nemotron 3.5 Super VL at 120 billion parameters and Nemotron 3.5 Bolt, a 74B-A9B model designed to fit a 24GB GPU. The company also says an E4B version designed for a 4GB footprint is due October 23. That spread matters more to builders than a single headline model size: it maps to different hardware and latency budgets.
Match the claim to your workload
NVIDIA reports up to five times the throughput of comparable models for Bolt. Treat that as a vendor claim until it is reproduced on your own hardware and task. For a useful comparison, hold prompts, output limits, quantization and hardware constant, then test the actual visual or agent workflow you plan to ship.
A small model can make local deployment practical, while a larger vision model may suit richer image tasks. The right choice depends on quality under your constraints, not parameter count alone.
Source published 2026-10-15. Coverage is based on the maker’s announcement and demonstration.