{
  "id": 20297009,
  "title": "NVIDIA Nemotron 3.5 Lightning: The 30B Agent Model That Fits on One GPU",
  "url": "https://urgent.news/2026/11/16/nvidia-nemotron-3-5-lightning-the-30b-agent-model-that-fits-on-one-gpu",
  "topic": "tech",
  "section": "Tech",
  "published": "2026-11-16T15:38:16.000Z",
  "source": {
    "name": "KahawaTungu",
    "slug": "kahawatungu",
    "url": "https://kahawatungu.com/nvidia-nemotron-3-5-lightning-the-30b-agent-model-that-fits-on-one-gpu/"
  },
  "original_language": "en",
  "account": "On August 11, 2026, NVIDIA introduced Nemotron 3.5 Lightning, a 30-billion-parameter model designed to efficiently run long-running AI agents without requiring a multi-GPU cluster. This hybrid Mixture-of-Experts model interleaves Mamba-2 state-space layers with MoE and Attention layers, allowing it to maintain low inference costs while handling reasoning and long-range retrieval tasks. With a context length of up to 1 million tokens, Nemotron 3.5 Lightning supports multiple languages and can be deployed on a single 80GB H100 or A100 GPU, or even a GeForce RTX 5090, making it accessible for businesses with existing hardware. The model's performance, measured on benchmarks like MMLU Pro and SWE-bench, outperforms many other 30B models, and its open license permits unrestricted commercial use.",
  "summary": "NVIDIA released Nemotron 3.5 Lightning on August 11, 2026, and it arrives with a clear purpose: to serve as the high-volume execution layer for long-running AI agents without demanding a multi-GPU cluster to run. It is a 30-billion-parameter model that activates only 3 billion parameters per token, built for developers who need speed, long context, [...] The post NVIDIA Nemotron 3.5 Lightning:…",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}