{
 "id": "qwen3.8-2.4t-a95b",
 "name": "Qwen3.8 2.4T-A95B",
 "repo": "Qwen/Qwen3.8-2.4T-A95B",
 "lab": {
  "name": "Alibaba",
  "country": "China"
 },
 "released": "2026-08-08",
 "parameters": 2446.182725504,
 "activeParameters": 95,
 "maxContext": 262144,
 "architecture": "92 layers · 23 full attention (4 KV × 256) · 69 linear attention",
 "page": "https://studiotvai.com/vram-requirements/qwen3-8-2-4t-a95b",
 "calculator": "https://studiotvai.com/llm-vram-calculator?model=qwen3.8-2.4t-a95b",
 "weightsGiB": {
  "bf16": 4507.24,
  "fp8": 2257.41,
  "q4_k_m": 1361.03
 },
 "kvCacheGiBPerRequest": [
  {
   "context": 8192,
   "bf16": 0.719,
   "fp8": 0.359
  },
  {
   "context": 32768,
   "bf16": 2.875,
   "fp8": 1.438
  },
  {
   "context": 131072,
   "bf16": 11.5,
   "fp8": 5.75
  },
  {
   "context": 262144,
   "bf16": 23,
   "fp8": 11.5
  }
 ],
 "recurrentStateGiBPerRequest": 0.547,
 "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": null,
     "fp8": 128,
     "q4_k_m": 96
    }
   },
   {
    "gpu": "rtx-5090",
    "name": "GeForce RTX 5090",
    "memoryGB": 32,
    "engine": "llama.cpp",
    "byFormat": {
     "bf16": null,
     "fp8": 96,
     "q4_k_m": 64
    }
   },
   {
    "gpu": "rtx-pro-6000",
    "name": "RTX PRO 6000 Blackwell",
    "memoryGB": 96,
    "engine": "llama.cpp",
    "byFormat": {
     "bf16": 64,
     "fp8": 32,
     "q4_k_m": 16
    }
   },
   {
    "gpu": "mac-m5-max-128",
    "name": "Mac M5 Max 128 GB",
    "memoryGB": 96,
    "engine": "llama.cpp",
    "byFormat": {
     "bf16": 64,
     "fp8": 32,
     "q4_k_m": 16
    }
   },
   {
    "gpu": "h100",
    "name": "NVIDIA H100",
    "memoryGB": 80,
    "engine": "vLLM",
    "byFormat": {
     "bf16": 96,
     "fp8": 48,
     "q4_k_m": 24
    }
   },
   {
    "gpu": "h200",
    "name": "NVIDIA H200",
    "memoryGB": 141,
    "engine": "vLLM",
    "byFormat": {
     "bf16": 48,
     "fp8": 24,
     "q4_k_m": 16
    }
   },
   {
    "gpu": "b200",
    "name": "NVIDIA B200 (HGX)",
    "memoryGB": 180,
    "engine": "vLLM",
    "byFormat": {
     "bf16": 32,
     "fp8": 16,
     "q4_k_m": 16
    }
   }
  ]
 },
 "ggufFiles": null,
 "runsWith": [],
 "verified": null,
 "updated": "2026-10-10T15:05:26.095Z"
}