SYLIN

Nemotron 3.5 Lightning

by NVIDIA · Model · Language & code models

NVIDIA released Nemotron 3.5 Lightning, a 30B mixture-of-experts model with 3B active parameters per token

AccessOpen weightsNeeds~75 GBSize30B parametersLicenseOpenMDW 1.1

Evidence

  1. Aug 12
    Served with vLLM using FlashInfer Mamba backend, FP16 Mamba state and stochastic rounding; used close to 75 GB VRAM, reducible via KV cache
  2. Aug 11
    Per NVIDIA's chart, does not win coding or terminal benchmarks against same-size peers; wins instruction following and long context
  3. Aug 11
    Sits near the bottom of the intelligence index, well behind Opus and Kimi K3 frontier models

Cost to run

Alternatives

Maxwell FlashClaimed at or near the top on reasoning and leading on agentic search, even passing proprietary models