NVIDIA Expands Nemotron Family with Efficient Super VL and Edge‑Optimized Bolt Models

NVIDIA has added three new open models to the Nemotron lineup: the 120‑billion‑parameter Nemotron 3.5 Super VL, the Bolt 74B‑A9B 24GB variant, and the Bolt E4B 4GB version. Super VL targets always‑on agents that need high accuracy while staying compact, delivering performance comparable to much larger reasoning models at roughly one‑fifth the parameter count.

Performance and Efficiency Gains

According to internal benchmarks, Super VL completes tasks up to five times faster than competing models in its class, making it suitable for high‑volume workflows that span days or weeks. Its open license lets organizations post‑train the model on proprietary data and run it on their own NVIDIA infrastructure without vendor lock‑in.

The Bolt family brings agentic intelligence to the edge. Bolt 74B‑A9B 24GB uses 2‑bit quantization to fit into a 24 GB frame buffer, allowing it to run alongside other applications on RTX/DGX Spark systems. Bolt E4B 4GB is tuned for conversational and physical agents on GeForce RTX, Jetson, and similar platforms, enabling fast local voice assistants and interactive characters.

• Super VL: 120 B parameters, frontier accuracy, 5× speedup.
• Bolt 74B‑A9B 24GB: balanced intelligence/speed for local agents.
• Bolt E4B 4GB: optimized for conversational edge use.

Why it matters for GPU / AI infrastructure: These models expand the range of deployable agentic workloads on NVIDIA‑based hardware, from data‑center servers to desktop and edge devices. Customers of AiGPU can now select an appropriately sized open model that matches their compute budget while retaining the ability to customize and run agents locally.