{
 "id": "ministral-3-8b",
 "name": "Ministral 3 8B",
 "repo": "mistralai/Ministral-3-8B-Instruct-2512",
 "lab": {
  "name": "Mistral AI",
  "country": "France"
 },
 "released": "2025-12-02",
 "parameters": 8.92,
 "activeParameters": null,
 "maxContext": 262144,
 "architecture": "34 layers · GQA, 8 KV heads × 128 · vision encoder",
 "page": "https://studiotvai.com/vram-requirements/ministral-3-8b",
 "calculator": "https://studiotvai.com/llm-vram-calculator?model=ministral-3-8b",
 "weightsGiB": {
  "native": 9.7,
  "bf16": 16.61,
  "fp8": 9.31,
  "q4_k_m": 4.88
 },
 "kvCacheGiBPerRequest": [
  {
   "context": 8192,
   "bf16": 1.063,
   "fp8": 0.531
  },
  {
   "context": 32768,
   "bf16": 4.25,
   "fp8": 2.125
  },
  {
   "context": 131072,
   "bf16": 17,
   "fp8": 8.5
  },
  {
   "context": 262144,
   "bf16": 34,
   "fp8": 17
  }
 ],
 "recurrentStateGiBPerRequest": 0,
 "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": 1,
     "bf16": 1,
     "fp8": 1,
     "q4_k_m": 1
    }
   },
   {
    "gpu": "rtx-5090",
    "name": "GeForce RTX 5090",
    "memoryGB": 32,
    "engine": "llama.cpp",
    "byFormat": {
     "native": 1,
     "bf16": 1,
     "fp8": 1,
     "q4_k_m": 1
    }
   },
   {
    "gpu": "rtx-pro-6000",
    "name": "RTX PRO 6000 Blackwell",
    "memoryGB": 96,
    "engine": "llama.cpp",
    "byFormat": {
     "native": 1,
     "bf16": 1,
     "fp8": 1,
     "q4_k_m": 1
    }
   },
   {
    "gpu": "mac-m5-max-128",
    "name": "Mac M5 Max 128 GB",
    "memoryGB": 96,
    "engine": "llama.cpp",
    "byFormat": {
     "native": 1,
     "bf16": 1,
     "fp8": 1,
     "q4_k_m": 1
    }
   },
   {
    "gpu": "strix-halo",
    "name": "Ryzen AI Max+ 395 128 GB",
    "memoryGB": 96,
    "engine": "llama.cpp",
    "byFormat": {
     "native": 1,
     "bf16": 1,
     "fp8": 1,
     "q4_k_m": 1
    }
   },
   {
    "gpu": "dgx-spark",
    "name": "DGX Spark (GB10)",
    "memoryGB": 128,
    "engine": "llama.cpp",
    "byFormat": {
     "native": 1,
     "bf16": 1,
     "fp8": 1,
     "q4_k_m": 1
    }
   },
   {
    "gpu": "h100",
    "name": "NVIDIA H100",
    "memoryGB": 80,
    "engine": "vLLM",
    "byFormat": {
     "native": 1,
     "bf16": 1,
     "fp8": 1,
     "q4_k_m": 1
    }
   },
   {
    "gpu": "h200",
    "name": "NVIDIA H200",
    "memoryGB": 141,
    "engine": "vLLM",
    "byFormat": {
     "native": 1,
     "bf16": 1,
     "fp8": 1,
     "q4_k_m": 1
    }
   },
   {
    "gpu": "b200",
    "name": "NVIDIA B200 (HGX)",
    "memoryGB": 180,
    "engine": "vLLM",
    "byFormat": {
     "native": 1,
     "bf16": 1,
     "fp8": 1,
     "q4_k_m": 1
    }
   }
  ]
 },
 "ggufFiles": {
  "repo": "unsloth/Ministral-3-8B-Instruct-2512-GGUF",
  "gib": {
   "BF16": 15.82,
   "IQ4_NL": 4.6,
   "IQ4_XS": 4.39,
   "Q2_K": 3.12,
   "Q2_K_L": 3.24,
   "Q3_K_M": 3.95,
   "Q3_K_S": 3.6,
   "Q4_0": 4.6,
   "Q4_1": 5.05,
   "Q4_K_M": 4.84,
   "Q4_K_S": 4.61,
   "Q5_K_M": 5.64,
   "Q5_K_S": 5.51,
   "Q6_K": 6.49,
   "Q8_0": 8.41,
   "UD-IQ1_M": 2.25,
   "UD-IQ1_S": 2.12,
   "UD-IQ2_M": 2.95,
   "UD-IQ2_XXS": 2.46,
   "UD-IQ3_XXS": 3.26,
   "UD-Q2_K_XL": 3.32,
   "UD-Q3_K_XL": 4.13,
   "UD-Q4_K_XL": 4.92,
   "UD-Q5_K_XL": 5.66,
   "UD-Q6_K_XL": 7.19,
   "UD-Q8_K_XL": 10.33
  }
 },
 "runsWith": [
  "vLLM",
  "SGLang",
  "llama.cpp (Ollama, LM Studio)",
  "MLX (Mac)"
 ],
 "verified": {
  "gguf": {
   "day": "2026-10-30",
   "repo": "unsloth/Ministral-3-8B-Instruct-2512-GGUF",
   "kv": "match",
   "ours": 139264,
   "theirs": 139264
  }
 },
 "updated": "2026-10-30T00:07:05.915Z"
}