vLLM supports Nemotron 3.5 Super VL with a day-0 serving recipe

AISummary

vLLM says Nemotron 3.5 Super VL can be served from its v0.31.0 Docker image on day zero. The blog gives a 1x GB200 NVFP4 example with a 114,688-token maximum model length, the nemotron_v3 reasoning parser, and MTP with 5 speculative tokens. vLLM advises validating the configuration on the target hardware, and points to a cookbook from NVIDIA's Nemotron team.

Post on XView on X

vLLM

@vllm_project

Part of a thread · earlier post

Ready to serve Nemotron 3.5 Super VL?

Start with vllm/vllm-openai:v0.31.0. The blog includes a 1× GB200 NVFP4 example with --max-model-len 114688, --reasoning-parser nemotron_v3, and MTP with 5 speculative tokens. Validate the configuration on your target hardware.

Cookbook: https://github.com/NVIDIA-NeMo/Nemotron/blob/main/usage-cookbook/NVIDIA-Nemotron-3.5-Super-VL/vllm_cookbook.ipynb

Blog: https://vllm.ai/blog/2026-10-15-nemotron-3-5-super-vl-vllm

Thanks to the NVIDIA Nemotron team and the vLLM contributors for the Day-0 collaboration!

4/4

Source: vLLM · x.comPublished Oct 15, 11:50 AM ET · added here 12:00 PM ET

vLLM supports Nemotron 3.5 Super VL with a day-0 serving recipe

Topics