NVIDIA Launches Nemotron 3.5 Models

NVIDIA is pushing open models toward practical, long-running agents by pairing multimodal capability with aggressive inference efficiency.

• –Super VL combines reasoning, tool use, and text, image, document, and video understanding.
• –NVIDIA claims up to 5x higher throughput than comparable open models for agentic workloads.
• –Bolt 74B-A9B uses native 2-bit quantization to fit within 24GB and run locally alongside other applications.
• –Open weights, datasets, recipes, and support for vLLM, SGLang, Ollama, and llama.cpp improve customization and deployment flexibility.
• –The headline benchmark claims still need broader independent evaluation, especially for real-world agent reliability.