{
 "id": "minimax-m3",
 "name": "MiniMax M3",
 "repo": "MiniMaxAI/MiniMax-M3",
 "lab": {
  "name": "MiniMax",
  "country": "China"
 },
 "released": "2026-06-02",
 "parameters": 427.04,
 "activeParameters": 23,
 "maxContext": 1048576,
 "architecture": "MoE 128 experts + 1 shared · 60 layers, MiniMax Sparse Attention on 57 (GQA 4 KV × 128, 128-dim index keys) · vision encoder",
 "page": "https://studiotvai.com/vram-requirements/minimax-m3",
 "calculator": "https://studiotvai.com/llm-vram-calculator?model=minimax-m3",
 "weightsGiB": {
  "native": 413.27,
  "bf16": 795.42,
  "fp8": 400,
  "q4_k_m": 239.99
 },
 "kvCacheGiBPerRequest": [
  {
   "context": 8192,
   "bf16": 1.049,
   "fp8": 0.58
  },
  {
   "context": 32768,
   "bf16": 4.195,
   "fp8": 2.32
  },
  {
   "context": 131072,
   "bf16": 16.781,
   "fp8": 9.281
  },
  {
   "context": 1048576,
   "bf16": 134.25,
   "fp8": 74.25
  }
 ],
 "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": 24,
     "bf16": 48,
     "fp8": 24,
     "q4_k_m": 16
    }
   },
   {
    "gpu": "rtx-5090",
    "name": "GeForce RTX 5090",
    "memoryGB": 32,
    "engine": "llama.cpp",
    "byFormat": {
     "native": 16,
     "bf16": 32,
     "fp8": 16,
     "q4_k_m": 16
    }
   },
   {
    "gpu": "rtx-pro-6000",
    "name": "RTX PRO 6000 Blackwell",
    "memoryGB": 96,
    "engine": "llama.cpp",
    "byFormat": {
     "native": 5,
     "bf16": 16,
     "fp8": 5,
     "q4_k_m": 3
    }
   },
   {
    "gpu": "mac-m5-max-128",
    "name": "Mac M5 Max 128 GB",
    "memoryGB": 96,
    "engine": "llama.cpp",
    "byFormat": {
     "native": 5,
     "bf16": 16,
     "fp8": 5,
     "q4_k_m": 3
    }
   },
   {
    "gpu": "strix-halo",
    "name": "Ryzen AI Max+ 395 128 GB",
    "memoryGB": 96,
    "engine": "llama.cpp",
    "byFormat": {
     "native": 5,
     "bf16": 16,
     "fp8": 5,
     "q4_k_m": 3
    }
   },
   {
    "gpu": "dgx-spark",
    "name": "DGX Spark (GB10)",
    "memoryGB": 128,
    "engine": "llama.cpp",
    "byFormat": {
     "native": 4,
     "bf16": 7,
     "fp8": 4,
     "q4_k_m": 2
    }
   },
   {
    "gpu": "h100",
    "name": "NVIDIA H100",
    "memoryGB": 80,
    "engine": "vLLM",
    "byFormat": {
     "native": 7,
     "bf16": 16,
     "fp8": 7,
     "q4_k_m": 4
    }
   },
   {
    "gpu": "h200",
    "name": "NVIDIA H200",
    "memoryGB": 141,
    "engine": "vLLM",
    "byFormat": {
     "native": 4,
     "bf16": 7,
     "fp8": 4,
     "q4_k_m": 2
    }
   },
   {
    "gpu": "b200",
    "name": "NVIDIA B200 (HGX)",
    "memoryGB": 180,
    "engine": "vLLM",
    "byFormat": {
     "native": 3,
     "bf16": 6,
     "fp8": 3,
     "q4_k_m": 2
    }
   }
  ]
 },
 "ggufFiles": null,
 "runsWith": [],
 "verified": null,
 "updated": "2026-10-23T14:01:26.189Z"
}