{
 "id": "deepseek-v4-pro",
 "name": "DeepSeek V4 Pro",
 "repo": "deepseek-ai/DeepSeek-V4-Pro",
 "lab": {
  "name": "DeepSeek",
  "country": "China"
 },
 "released": "2026-04-22",
 "parameters": 1598.84,
 "activeParameters": 49,
 "maxContext": 1048576,
 "architecture": "MoE 384 experts · 512-wide latent: 128-token window + sequence compression (30 layers 4×, 31 layers 128×)",
 "page": "https://studiotvai.com/vram-requirements/deepseek-v4-pro",
 "calculator": "https://studiotvai.com/llm-vram-calculator?model=deepseek-v4-pro",
 "weightsGiB": {
  "native": 792.34,
  "bf16": 2929.94,
  "fp8": 1466.7,
  "q4_k_m": 884.66
 },
 "kvCacheGiBPerRequest": [
  {
   "context": 8192,
   "bf16": 0.075,
   "fp8": 0.046
  },
  {
   "context": 32768,
   "bf16": 0.28,
   "fp8": 0.172
  },
  {
   "context": 131072,
   "bf16": 1.096,
   "fp8": 0.677
  },
  {
   "context": 1048576,
   "bf16": 8.716,
   "fp8": 5.387
  }
 ],
 "recurrentStateGiBPerRequest": 0.0174,
 "gpusNeeded": {
  "scenario": "one request of 8192 tokens, BF16 KV cache; vLLM on data-center GPUs, llama.cpp elsewhere",
  "gpus": [
   {
    "gpu": "rtx-4090",
    "name": "GeForce RTX 4090",
    "memoryGB": 24,
    "engine": "llama.cpp",
    "byFormat": {
     "native": 48,
     "bf16": null,
     "fp8": 96,
     "q4_k_m": 48
    }
   },
   {
    "gpu": "rtx-5090",
    "name": "GeForce RTX 5090",
    "memoryGB": 32,
    "engine": "llama.cpp",
    "byFormat": {
     "native": 32,
     "bf16": 128,
     "fp8": 64,
     "q4_k_m": 32
    }
   },
   {
    "gpu": "rtx-pro-6000",
    "name": "RTX PRO 6000 Blackwell",
    "memoryGB": 96,
    "engine": "llama.cpp",
    "byFormat": {
     "native": 16,
     "bf16": 48,
     "fp8": 24,
     "q4_k_m": 16
    }
   },
   {
    "gpu": "mac-m5-max-128",
    "name": "Mac M5 Max 128 GB",
    "memoryGB": 96,
    "engine": "llama.cpp",
    "byFormat": {
     "native": 16,
     "bf16": 48,
     "fp8": 24,
     "q4_k_m": 16
    }
   },
   {
    "gpu": "h100",
    "name": "NVIDIA H100",
    "memoryGB": 80,
    "engine": "vLLM",
    "byFormat": {
     "native": 16,
     "bf16": 48,
     "fp8": 24,
     "q4_k_m": 16
    }
   },
   {
    "gpu": "h200",
    "name": "NVIDIA H200",
    "memoryGB": 141,
    "engine": "vLLM",
    "byFormat": {
     "native": 7,
     "bf16": 24,
     "fp8": 16,
     "q4_k_m": 8
    }
   },
   {
    "gpu": "b200",
    "name": "NVIDIA B200 (HGX)",
    "memoryGB": 180,
    "engine": "vLLM",
    "byFormat": {
     "native": 6,
     "bf16": 24,
     "fp8": 16,
     "q4_k_m": 6
    }
   }
  ]
 },
 "ggufFiles": {
  "repo": "DevQuasar/deepseek-ai.DeepSeek-V4-Pro-GGUF",
  "gib": {
   "Q2_K": 530.31,
   "Q3_K_M": 697.01,
   "Q4_K_M": 885.58,
   "Q5_K_M": 1038.96
  }
 },
 "runsWith": [
  "vLLM",
  "SGLang",
  "llama.cpp (Ollama, LM Studio)",
  "MLX (Mac)"
 ],
 "verified": null,
 "updated": "2026-10-10T15:05:26.095Z"
}