{
 "id": "inkling",
 "name": "Inkling",
 "repo": "thinkingmachines/Inkling",
 "lab": {
  "name": "Thinking Machines Lab",
  "country": "United States"
 },
 "released": "2026-07-15",
 "parameters": 952.38,
 "activeParameters": 41,
 "maxContext": 1048576,
 "architecture": "MoE 256 experts + 2 shared · 11 global layers (8 KV × 128) + 55 sliding-window layers (512 tokens, 16 KV × 128) · short convolutions · image and audio input",
 "page": "https://studiotvai.com/vram-requirements/inkling",
 "calculator": "https://studiotvai.com/llm-vram-calculator?model=inkling",
 "weightsGiB": {
  "native": 541.55,
  "bf16": 1764.15,
  "fp8": 884.37,
  "q4_k_m": 532.75
 },
 "kvCacheGiBPerRequest": [
  {
   "context": 8192,
   "bf16": 0.559,
   "fp8": 0.279
  },
  {
   "context": 32768,
   "bf16": 1.59,
   "fp8": 0.795
  },
  {
   "context": 131072,
   "bf16": 5.715,
   "fp8": 2.857
  },
  {
   "context": 1048576,
   "bf16": 44.215,
   "fp8": 22.107
  }
 ],
 "recurrentStateGiBPerRequest": 0.0081,
 "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": 32,
     "bf16": 96,
     "fp8": 48,
     "q4_k_m": 32
    }
   },
   {
    "gpu": "rtx-5090",
    "name": "GeForce RTX 5090",
    "memoryGB": 32,
    "engine": "llama.cpp",
    "byFormat": {
     "native": 24,
     "bf16": 64,
     "fp8": 32,
     "q4_k_m": 24
    }
   },
   {
    "gpu": "rtx-pro-6000",
    "name": "RTX PRO 6000 Blackwell",
    "memoryGB": 96,
    "engine": "llama.cpp",
    "byFormat": {
     "native": 6,
     "bf16": 24,
     "fp8": 16,
     "q4_k_m": 6
    }
   },
   {
    "gpu": "mac-m5-max-128",
    "name": "Mac M5 Max 128 GB",
    "memoryGB": 96,
    "engine": "llama.cpp",
    "byFormat": {
     "native": 6,
     "bf16": 24,
     "fp8": 16,
     "q4_k_m": 6
    }
   },
   {
    "gpu": "strix-halo",
    "name": "Ryzen AI Max+ 395 128 GB",
    "memoryGB": 96,
    "engine": "llama.cpp",
    "byFormat": {
     "native": 6,
     "bf16": 24,
     "fp8": 16,
     "q4_k_m": 6
    }
   },
   {
    "gpu": "dgx-spark",
    "name": "DGX Spark (GB10)",
    "memoryGB": 128,
    "engine": "llama.cpp",
    "byFormat": {
     "native": 5,
     "bf16": 16,
     "fp8": 8,
     "q4_k_m": 5
    }
   },
   {
    "gpu": "h100",
    "name": "NVIDIA H100",
    "memoryGB": 80,
    "engine": "vLLM",
    "byFormat": {
     "native": 16,
     "bf16": 32,
     "fp8": 16,
     "q4_k_m": 8
    }
   },
   {
    "gpu": "h200",
    "name": "NVIDIA H200",
    "memoryGB": 141,
    "engine": "vLLM",
    "byFormat": {
     "native": 5,
     "bf16": 16,
     "fp8": 8,
     "q4_k_m": 5
    }
   },
   {
    "gpu": "b200",
    "name": "NVIDIA B200 (HGX)",
    "memoryGB": 180,
    "engine": "vLLM",
    "byFormat": {
     "native": 4,
     "bf16": 16,
     "fp8": 6,
     "q4_k_m": 4
    }
   }
  ]
 },
 "ggufFiles": {
  "repo": "DevQuasar/thinkingmachines.Inkling-GGUF",
  "gib": {
   "IQ2_XXS": 229.54,
   "Q2_K": 320.33,
   "Q3_K_M": 420.06,
   "Q4_K_M": 533.18
  }
 },
 "runsWith": [
  "vLLM",
  "SGLang",
  "MLX (Mac)"
 ],
 "verified": null,
 "updated": "2026-10-30T00:07:05.915Z"
}