vLLM Adds Day 0 Support for NVIDIA Nemotron 3.5 Super VL
vLLM provides immediate serving support for NVIDIA's new Nemotron 3.5 Super VL, enabling developers to deploy the multimodal agent model via standard APIs.
ImportanceMaterialEvidenceE2 unreplicatedWrite-upQuick
vLLM announced "Day 0" support for NVIDIA's newly released Nemotron 3.5 Super VL model on October 15, 2026. This allows developers to run this specialized multimodal model directly within vLLM environments without waiting for third-party integrations.
Nemotron 3.5 Super VL is a mixture-of-experts (MoE) model with 120 billion total parameters and 12 billion active parameters, designed for long-running agents that require understanding of text, images, documents, and video. The vLLM support includes continuous batching, prefix caching, speculative decoding, and an OpenAI-compatible API, facilitating seamless integration into existing production workflows.
The update also supports Multi-Token Prediction (MTP) to accelerate generation. NVIDIA claims the model delivers up to 5x faster throughput for agentic tasks compared to other open models in its size category on a single GPU, though these figures are vendor-reported and have not been independently reproduced.