{
 "id": "deepseek-v4-flash-vision-exp",
 "name": "DeepSeek V4 Flash Vision Exp",
 "repo": "deepseek-ai/DeepSeek-V4-Flash-Vision-Exp",
 "lab": {
  "name": "DeepSeek",
  "country": "China"
 },
 "released": "2026-08-31",
 "parameters": 304.646824126,
 "activeParameters": 15,
 "maxContext": 1048576,
 "architecture": "43 layers · 512-wide latent, 128-token window + compression (21 layers 4×, 20 layers 128×)",
 "page": "https://studiotvai.com/vram-requirements/deepseek-v4-flash-vision-exp",
 "calculator": "https://studiotvai.com/llm-vram-calculator?model=deepseek-v4-flash-vision-exp",
 "weightsGiB": {
  "native": 146.18,
  "bf16": 530.48,
  "fp8": 266.23,
  "q4_k_m": 160.25
 },
 "kvCacheGiBPerRequest": [
  {
   "context": 8192,
   "bf16": 0.053,
   "fp8": 0.032
  },
  {
   "context": 32768,
   "bf16": 0.195,
   "fp8": 0.12
  },
  {
   "context": 131072,
   "bf16": 0.766,
   "fp8": 0.473
  },
  {
   "context": 1048576,
   "bf16": 6.088,
   "fp8": 3.763
  }
 ],
 "recurrentStateGiBPerRequest": 0.0114,
 "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": 7,
     "bf16": 32,
     "fp8": 16,
     "q4_k_m": 8
    }
   },
   {
    "gpu": "rtx-5090",
    "name": "GeForce RTX 5090",
    "memoryGB": 32,
    "engine": "llama.cpp",
    "byFormat": {
     "native": 6,
     "bf16": 24,
     "fp8": 16,
     "q4_k_m": 6
    }
   },
   {
    "gpu": "rtx-pro-6000",
    "name": "RTX PRO 6000 Blackwell",
    "memoryGB": 96,
    "engine": "llama.cpp",
    "byFormat": {
     "native": 2,
     "bf16": 6,
     "fp8": 3,
     "q4_k_m": 2
    }
   },
   {
    "gpu": "mac-m5-max-128",
    "name": "Mac M5 Max 128 GB",
    "memoryGB": 96,
    "engine": "llama.cpp",
    "byFormat": {
     "native": 2,
     "bf16": 6,
     "fp8": 3,
     "q4_k_m": 2
    }
   },
   {
    "gpu": "h100",
    "name": "NVIDIA H100",
    "memoryGB": 80,
    "engine": "vLLM",
    "byFormat": {
     "native": 3,
     "bf16": 8,
     "fp8": 4,
     "q4_k_m": 3
    }
   },
   {
    "gpu": "h200",
    "name": "NVIDIA H200",
    "memoryGB": 141,
    "engine": "vLLM",
    "byFormat": {
     "native": 2,
     "bf16": 5,
     "fp8": 3,
     "q4_k_m": 2
    }
   },
   {
    "gpu": "b200",
    "name": "NVIDIA B200 (HGX)",
    "memoryGB": 180,
    "engine": "vLLM",
    "byFormat": {
     "native": 1,
     "bf16": 4,
     "fp8": 2,
     "q4_k_m": 2
    }
   }
  ]
 },
 "ggufFiles": {
  "repo": "unsloth/DeepSeek-V4-Flash-Vision-Exp-GGUF",
  "gib": {
   "UD-IQ1_M": 80.9,
   "UD-IQ1_S": 76.78,
   "UD-IQ2_XXS": 84.46,
   "UD-IQ3_S": 106.54,
   "UD-IQ3_XXS": 95.93,
   "UD-IQ4_XS": 127.28,
   "UD-Q2_K_XL": 90.18,
   "UD-Q3_K_XL": 119.4,
   "UD-Q4_K_XL": 144.44,
   "UD-Q8_K_XL": 150.75
  }
 },
 "runsWith": [
  "vLLM",
  "SGLang",
  "llama.cpp (Ollama, LM Studio)",
  "MLX (Mac)"
 ],
 "verified": null,
 "updated": "2026-10-10T15:05:26.095Z"
}