SYLIN

Nemotron Super

by NVIDIA · Model · Language & code models

Nvidia Nemotron 3.5 Super VL: multimodal 120B MoE (12B active), over 600 tokens/s, open weights with training recipe, about 75 GB

AccessOpen weightsNeeds~75 GBDownload75 GBLicenseOpenMDW License 1.1

Evidence

  1. Oct 18
    Less intelligent than MiniMax or Qwen 3.8 27B but much faster
  2. Sep 3
    Cited as the only other prominent fully open stack (weights plus training materials)
  3. Aug 16
    Open 30B MoE with 1M context; below Qwen 3.6, near Qwen 3.5 and Gemma 4, but fastest in its size class; FP4 about 22 GB

Cost to run

Alternatives

PhoneLLMShown calling create_reservation where base Nemotron 3 Nano only talks and never books the table
Built on Nemotron Super