{
 "id": "mellum2.1-12b-a2.5b",
 "name": "Mellum2.1 12B-A2.5B",
 "repo": "JetBrains/Mellum2.1-12B-A2.5B-Thinking",
 "lab": {
  "name": "JetBrains",
  "country": "Czechia"
 },
 "released": "2026-09-20",
 "parameters": 12.149923072,
 "activeParameters": 2.5,
 "maxContext": 131072,
 "architecture": "28 layers · 7 full attention (4 KV × 128) · 21 sliding-window layers (1024 tokens, 4 KV × 128)",
 "page": "https://studiotvai.com/vram-requirements/mellum2-1-12b-a2-5b",
 "calculator": "https://studiotvai.com/llm-vram-calculator?model=mellum2.1-12b-a2.5b",
 "weightsGiB": {
  "bf16": 22.63,
  "fp8": 11.74,
  "q4_k_m": 7.37
 },
 "kvCacheGiBPerRequest": [
  {
   "context": 8192,
   "bf16": 0.15,
   "fp8": 0.075
  },
  {
   "context": 32768,
   "bf16": 0.479,
   "fp8": 0.239
  },
  {
   "context": 131072,
   "bf16": 1.791,
   "fp8": 0.896
  }
 ],
 "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": 1,
     "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": "bartowski/Mellum2.1-12B-A2.5B-Thinking-GGUF",
  "gib": {
   "IQ2_M": 5.03,
   "IQ3_M": 6.75,
   "IQ3_XS": 5.86,
   "IQ3_XXS": 5.85,
   "IQ4_NL": 7.14,
   "IQ4_XS": 6.49,
   "Q2_K": 5.15,
   "Q3_K_L": 6.64,
   "Q3_K_M": 6.24,
   "Q3_K_S": 5.64,
   "Q4_0": 6.52,
   "Q4_1": 7.18,
   "Q4_K_L": 8.59,
   "Q4_K_M": 8.21,
   "Q4_K_S": 7.32,
   "Q5_K_M": 9.28,
   "Q5_K_S": 8.35,
   "Q6_K": 10.67,
   "Q6_K_L": 10.99,
   "Q6_K_S": 10.37,
   "Q8_0": 12.04,
   "BF16": 22.64
  }
 },
 "runsWith": [
  "vLLM",
  "SGLang",
  "llama.cpp (Ollama, LM Studio)",
  "MLX (Mac)"
 ],
 "verified": null,
 "updated": "2026-10-10T15:05:26.095Z"
}