{
 "id": "gemma-4-26b-a4b",
 "name": "Gemma 4 26B-A4B",
 "repo": "google/gemma-4-26B-A4B-it",
 "lab": {
  "name": "Google",
  "country": "United States"
 },
 "released": "2026-03-11",
 "parameters": 25.81,
 "activeParameters": 4,
 "maxContext": 262144,
 "architecture": "MoE 128 experts · 25 sliding-window layers (1024 tokens) · 5 global layers (2 KV × 512)",
 "page": "https://studiotvai.com/vram-requirements/gemma-4-26b-a4b",
 "calculator": "https://studiotvai.com/llm-vram-calculator?model=gemma-4-26b-a4b",
 "weightsGiB": {
  "bf16": 48.07,
  "fp8": 24.72,
  "q4_k_m": 15.38
 },
 "kvCacheGiBPerRequest": [
  {
   "context": 8192,
   "bf16": 0.352,
   "fp8": 0.176
  },
  {
   "context": 32768,
   "bf16": 0.82,
   "fp8": 0.41
  },
  {
   "context": 131072,
   "bf16": 2.695,
   "fp8": 1.348
  },
  {
   "context": 262144,
   "bf16": 5.195,
   "fp8": 2.598
  }
 ],
 "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": 3,
     "fp8": 2,
     "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": "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": {
  "repo": "unsloth/gemma-4-26B-A4B-it-GGUF",
  "gib": {
   "BF16": 47.04,
   "MXFP4_MOE": 15.41,
   "Q8_0": 25.02,
   "UD-IQ2_M": 9.33,
   "UD-IQ2_XXS": 9.24,
   "UD-IQ3_S": 10.51,
   "UD-IQ3_XXS": 10.63,
   "UD-IQ4_NL": 12.68,
   "UD-IQ4_XS": 12.66,
   "UD-Q2_K_XL": 9.82,
   "UD-Q3_K_M": 11.85,
   "UD-Q3_K_XL": 12.02,
   "UD-Q4_K_M": 15.78,
   "UD-Q4_K_S": 15.36,
   "UD-Q4_K_XL": 15.84,
   "UD-Q5_K_M": 19.7,
   "UD-Q5_K_S": 17.56,
   "UD-Q5_K_XL": 19.76,
   "UD-Q6_K": 21.58,
   "UD-Q6_K_XL": 21.7,
   "UD-Q8_K_XL": 25.74
  }
 },
 "runsWith": [
  "vLLM",
  "SGLang",
  "llama.cpp (Ollama, LM Studio)",
  "MLX (Mac)"
 ],
 "verified": null,
 "updated": "2026-10-10T15:05:26.095Z"
}