NVIDIA: Nemotron 3.5 Super VL

NVIDIA's Nemotron 3.5 Super VL accepts text, image, and video input and returns text, with a context window of 1,048,576 tokens and a maximum completion length of 235,929 tokens. It supports tool calling and a reasoning mode, which suits agent workflows that need to inspect visual material or long documents before acting. The available facts do not confirm structured output support, and they do not say whether the weights are open, so teams that depend on either should verify before committing. At $0.10 per million input tokens and $0.40 per million output tokens, this model belongs on the shortlist for cost-sensitive multimodal workloads, especially those that feed large volumes of input into a long context. Its evidence base is thin, however. The blended score of 40.1 rests on a single independent benchmark, an AI index reading of 43.7, so its standing is largely unproven. Buyers should run their own evaluations against alternatives before relying on it.