{
 "id": "kimi-vl-a3b-instruct",
 "name": "Kimi VL A3B Instruct",
 "repo": "moonshotai/Kimi-VL-A3B-Instruct",
 "lab": {
  "name": "Moonshot AI",
  "country": "China"
 },
 "released": "2025-04-09",
 "parameters": 16.41,
 "activeParameters": 3,
 "maxContext": 131072,
 "architecture": "MoE 64 experts + 2 shared · MLA (512 + 64 latent) on 27 layers · MoonViT vision encoder",
 "page": "https://studiotvai.com/vram-requirements/kimi-vl-a3b-instruct",
 "calculator": "https://studiotvai.com/llm-vram-calculator?model=kimi-vl-a3b-instruct",
 "weightsGiB": {
  "bf16": 30.57,
  "fp8": 15.91,
  "q4_k_m": 9.69
 },
 "kvCacheGiBPerRequest": [
  {
   "context": 8192,
   "bf16": 0.237,
   "fp8": 0.119
  },
  {
   "context": 32768,
   "bf16": 0.949,
   "fp8": 0.475
  },
  {
   "context": 131072,
   "bf16": 3.797,
   "fp8": 1.898
  }
 ],
 "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": {
     "bf16": 2,
     "fp8": 1,
     "q4_k_m": 1
    }
   },
   {
    "gpu": "rtx-5090",
    "name": "GeForce RTX 5090",
    "memoryGB": 32,
    "engine": "llama.cpp",
    "byFormat": {
     "bf16": 2,
     "fp8": 1,
     "q4_k_m": 1
    }
   },
   {
    "gpu": "rtx-pro-6000",
    "name": "RTX PRO 6000 Blackwell",
    "memoryGB": 96,
    "engine": "llama.cpp",
    "byFormat": {
     "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": {
     "bf16": 1,
     "fp8": 1,
     "q4_k_m": 1
    }
   },
   {
    "gpu": "strix-halo",
    "name": "Ryzen AI Max+ 395 128 GB",
    "memoryGB": 96,
    "engine": "llama.cpp",
    "byFormat": {
     "bf16": 1,
     "fp8": 1,
     "q4_k_m": 1
    }
   },
   {
    "gpu": "dgx-spark",
    "name": "DGX Spark (GB10)",
    "memoryGB": 128,
    "engine": "llama.cpp",
    "byFormat": {
     "bf16": 1,
     "fp8": 1,
     "q4_k_m": 1
    }
   },
   {
    "gpu": "h100",
    "name": "NVIDIA H100",
    "memoryGB": 80,
    "engine": "vLLM",
    "byFormat": {
     "bf16": 1,
     "fp8": 1,
     "q4_k_m": 1
    }
   },
   {
    "gpu": "h200",
    "name": "NVIDIA H200",
    "memoryGB": 141,
    "engine": "vLLM",
    "byFormat": {
     "bf16": 1,
     "fp8": 1,
     "q4_k_m": 1
    }
   },
   {
    "gpu": "b200",
    "name": "NVIDIA B200 (HGX)",
    "memoryGB": 180,
    "engine": "vLLM",
    "byFormat": {
     "bf16": 1,
     "fp8": 1,
     "q4_k_m": 1
    }
   }
  ]
 },
 "ggufFiles": null,
 "runsWith": [],
 "verified": null,
 "updated": "2026-10-25T11:03:26.246Z"
}