{
 "license": "CC BY 4.0",
 "attribution": "Hundredth Run Compute Buyer Desk, https://hundredthrun-compute-desk.pages.dev/",
 "overhead": 1.2,
 "method": "VRAM GB = parameters in billions x bytes per parameter x overhead. A rule of thumb, not a benchmark.",
 "models": [
  {
   "slug": "llama-3-1-8b",
   "name": "Llama 3.1 8B",
   "params_b": 8.0,
   "architecture": "dense",
   "source_url": "https://huggingface.co/meta-llama/Llama-3.1-8B",
   "date_checked": "2026-11-04",
   "vram_gb": {
    "fp16": 19.2,
    "fp8": 9.6,
    "int4": 4.8,
    "lora": 19.2,
    "qlora": 4.8,
    "full": 153.6
   }
  },
  {
   "slug": "llama-3-1-70b",
   "name": "Llama 3.1 70B",
   "params_b": 70.6,
   "architecture": "dense",
   "source_url": "https://huggingface.co/meta-llama/Llama-3.1-70B",
   "date_checked": "2026-11-04",
   "vram_gb": {
    "fp16": 169.4,
    "fp8": 84.7,
    "int4": 42.4,
    "lora": 169.4,
    "qlora": 42.4,
    "full": 1355.5
   }
  },
  {
   "slug": "llama-3-3-70b",
   "name": "Llama 3.3 70B",
   "params_b": 70.6,
   "architecture": "dense",
   "source_url": "https://huggingface.co/meta-llama/Llama-3.3-70B-Instruct",
   "date_checked": "2026-11-04",
   "vram_gb": {
    "fp16": 169.4,
    "fp8": 84.7,
    "int4": 42.4,
    "lora": 169.4,
    "qlora": 42.4,
    "full": 1355.5
   }
  },
  {
   "slug": "llama-3-1-405b",
   "name": "Llama 3.1 405B",
   "params_b": 405.9,
   "architecture": "dense",
   "source_url": "https://huggingface.co/meta-llama/Llama-3.1-405B",
   "date_checked": "2026-11-04",
   "vram_gb": {
    "fp16": 974.2,
    "fp8": 487.1,
    "int4": 243.5,
    "lora": 974.2,
    "qlora": 243.5,
    "full": 7793.3
   }
  },
  {
   "slug": "llama-4-scout",
   "name": "Llama 4 Scout",
   "params_b": 108.6,
   "architecture": "moe",
   "source_url": "https://huggingface.co/meta-llama/Llama-4-Scout-17B-16E",
   "date_checked": "2026-11-04",
   "vram_gb": {
    "fp16": 260.6,
    "fp8": 130.3,
    "int4": 65.2,
    "lora": 260.6,
    "qlora": 65.2,
    "full": 2085.1
   }
  },
  {
   "slug": "mistral-7b",
   "name": "Mistral 7B",
   "params_b": 7.2,
   "architecture": "dense",
   "source_url": "https://huggingface.co/mistralai/Mistral-7B-v0.3",
   "date_checked": "2026-11-04",
   "vram_gb": {
    "fp16": 17.3,
    "fp8": 8.6,
    "int4": 4.3,
    "lora": 17.3,
    "qlora": 4.3,
    "full": 138.2
   }
  },
  {
   "slug": "mixtral-8x7b",
   "name": "Mixtral 8x7B",
   "params_b": 46.7,
   "architecture": "moe",
   "source_url": "https://huggingface.co/mistralai/Mixtral-8x7B-v0.1",
   "date_checked": "2026-11-04",
   "vram_gb": {
    "fp16": 112.1,
    "fp8": 56.0,
    "int4": 28.0,
    "lora": 112.1,
    "qlora": 28.0,
    "full": 896.6
   }
  },
  {
   "slug": "qwen2-5-7b",
   "name": "Qwen2.5 7B",
   "params_b": 7.6,
   "architecture": "dense",
   "source_url": "https://huggingface.co/Qwen/Qwen2.5-7B",
   "date_checked": "2026-11-04",
   "vram_gb": {
    "fp16": 18.2,
    "fp8": 9.1,
    "int4": 4.6,
    "lora": 18.2,
    "qlora": 4.6,
    "full": 145.9
   }
  },
  {
   "slug": "qwen2-5-32b",
   "name": "Qwen2.5 32B",
   "params_b": 32.8,
   "architecture": "dense",
   "source_url": "https://huggingface.co/Qwen/Qwen2.5-32B",
   "date_checked": "2026-11-04",
   "vram_gb": {
    "fp16": 78.7,
    "fp8": 39.4,
    "int4": 19.7,
    "lora": 78.7,
    "qlora": 19.7,
    "full": 629.8
   }
  },
  {
   "slug": "qwen2-5-72b",
   "name": "Qwen2.5 72B",
   "params_b": 72.7,
   "architecture": "dense",
   "source_url": "https://huggingface.co/Qwen/Qwen2.5-72B",
   "date_checked": "2026-11-04",
   "vram_gb": {
    "fp16": 174.5,
    "fp8": 87.2,
    "int4": 43.6,
    "lora": 174.5,
    "qlora": 43.6,
    "full": 1395.8
   }
  },
  {
   "slug": "qwen3-32b",
   "name": "Qwen3 32B",
   "params_b": 32.8,
   "architecture": "dense",
   "source_url": "https://huggingface.co/Qwen/Qwen3-32B",
   "date_checked": "2026-11-04",
   "vram_gb": {
    "fp16": 78.7,
    "fp8": 39.4,
    "int4": 19.7,
    "lora": 78.7,
    "qlora": 19.7,
    "full": 629.8
   }
  },
  {
   "slug": "qwen3-30b-a3b",
   "name": "Qwen3 30B-A3B",
   "params_b": 30.5,
   "architecture": "moe",
   "source_url": "https://huggingface.co/Qwen/Qwen3-30B-A3B",
   "date_checked": "2026-11-04",
   "vram_gb": {
    "fp16": 73.2,
    "fp8": 36.6,
    "int4": 18.3,
    "lora": 73.2,
    "qlora": 18.3,
    "full": 585.6
   }
  },
  {
   "slug": "gemma-2-27b",
   "name": "Gemma 2 27B",
   "params_b": 27.2,
   "architecture": "dense",
   "source_url": "https://huggingface.co/google/gemma-2-27b",
   "date_checked": "2026-11-04",
   "vram_gb": {
    "fp16": 65.3,
    "fp8": 32.6,
    "int4": 16.3,
    "lora": 65.3,
    "qlora": 16.3,
    "full": 522.2
   }
  },
  {
   "slug": "deepseek-v3",
   "name": "DeepSeek V3",
   "params_b": 684.5,
   "architecture": "moe",
   "source_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3",
   "date_checked": "2026-11-04",
   "vram_gb": {
    "fp16": 1642.8,
    "fp8": 821.4,
    "int4": 410.7,
    "lora": 1642.8,
    "qlora": 410.7,
    "full": 13142.4
   }
  },
  {
   "slug": "gpt-oss-120b",
   "name": "gpt-oss-120b",
   "params_b": 116.8,
   "architecture": "moe",
   "source_url": "https://huggingface.co/openai/gpt-oss-120b",
   "date_checked": "2026-11-04",
   "vram_gb": {
    "fp16": 280.3,
    "fp8": 140.2,
    "int4": 70.1,
    "lora": 280.3,
    "qlora": 70.1,
    "full": 2242.6
   }
  },
  {
   "slug": "gpt-oss-20b",
   "name": "gpt-oss-20b",
   "params_b": 20.9,
   "architecture": "moe",
   "source_url": "https://huggingface.co/openai/gpt-oss-20b",
   "date_checked": "2026-11-04",
   "vram_gb": {
    "fp16": 50.2,
    "fp8": 25.1,
    "int4": 12.5,
    "lora": 50.2,
    "qlora": 12.5,
    "full": 401.3
   }
  }
 ]
}