{
 "title": "LLM VRAM Requirements 2026: The Open-Model GPU Index",
 "url": "https://www.servnetuk.com/research/open-model-vram-reference-index-2026",
 "publisher": "Servnet Research",
 "license": "CC BY 4.0 — free to reuse with attribution",
 "collected": "2026-08-03",
 "records": [
  {
   "metric": "Llama 4 Scout VRAM at INT4",
   "value": "~55 GB",
   "source": "Spheron"
  },
  {
   "metric": "Llama 4 Scout VRAM at FP16",
   "value": "~218 GB",
   "source": "Spheron"
  },
  {
   "metric": "Llama 4 Maverick VRAM at INT4",
   "value": "~200 GB",
   "source": "Spheron"
  },
  {
   "metric": "DeepSeek V4 VRAM at FP8",
   "value": "~1,000 GB",
   "source": "Spheron"
  },
  {
   "metric": "Kimi K2.5 VRAM at INT4",
   "value": "~630 GB",
   "source": "Spheron"
  },
  {
   "metric": "8B model VRAM at Q4",
   "value": "~5 GB",
   "source": "LocalAIMaster"
  },
  {
   "metric": "14B-class model VRAM at Q4",
   "value": "~9-10 GB",
   "source": "LocalAIMaster"
  },
  {
   "metric": "32B dense model VRAM at Q4",
   "value": "~20 GB",
   "source": "LocalAIMaster"
  },
  {
   "metric": "70B-and-up model VRAM at Q4",
   "value": "40 GB+",
   "source": "LocalAIMaster"
  },
  {
   "metric": "KV-cache and activation overhead allowance",
   "value": "20-40%",
   "source": "Data Dynamics"
  }
 ]
}