vLLM adds Day-0 serving support for NVIDIA Nemotron 3.5 Super VL

vLLM announces Day-0 support for NVIDIA Nemotron 3.5 Super VL, an open multimodal model with 120B total parameters, 12B active per token, and a 1M-token context window.

The model accepts text, images, documents and video, and supports reasoning and tool use for agent workflows.

vLLM publishes a serving guide that starts from its v0.31.0 Docker image. Its 1x GB200 NVFP4 example uses a 114,688-token maximum model length, the nemotron_v3 reasoning parser, and multi-token prediction (MTP) with 5 speculative tokens. vLLM advises users to validate the configuration on their target hardware and points to a cookbook from NVIDIA's Nemotron team.