{
 "built": "2026-11-02",
 "fetched": {
  "date": "2026-11-02",
  "modelfit_updated": "2026-11-02",
  "modelfit_license": "CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/)"
 },
 "attribution": {
  "modelfit": {
   "name": "ModelFit",
   "url": "https://modelfit.io/",
   "license": "CC BY 4.0",
   "license_url": "https://creativecommons.org/licenses/by/4.0/",
   "dataset": "https://modelfit.io/data/",
   "updated": "2026-11-02"
  }
 },
 "unmatched_modelfit_ids": [
  "laguna-xs2-q4",
  "ling-3.0-flash-q4xs",
  "mellum2.1-12b-q4",
  "mistral-small-3.1-cloud",
  "nemotron-3-super-120b-q4xs",
  "nemotron-3.5-lightning-30b-q4xs",
  "nemotron-cascade-2-30b-a3b-q6",
  "north-mini-code-1-q4",
  "phi3-medium-q4",
  "phi3-mini-q4",
  "qwen3-235b-a22b-cloud",
  "qwen3-next-80b-a3b-q8",
  "qwen3.8-flash-next-q3",
  "qwen3.8-flash-next-q4xs"
 ],
 "models": [
  {
   "slug": "glm-5-3-flash",
   "name": "GLM-5.3-Flash",
   "maker": "Z.ai",
   "total_b": 320,
   "active_b": 18,
   "active_source": "card",
   "arch": "moe",
   "use": "Very large MoE",
   "source": null,
   "checked": "2026-09-14",
   "gpuwerk": "/docs/models/glm-5-3-flash/",
   "spark": null,
   "hf_check": null,
   "variants": [],
   "eci": null,
   "shortlist": true
  },
  {
   "slug": "deepseek-v4-flash",
   "name": "DeepSeek-V4-Flash",
   "maker": "DeepSeek",
   "total_b": 284,
   "active_b": 13,
   "active_source": "card",
   "arch": "moe",
   "use": "Very large fast MoE",
   "source": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash",
   "checked": "2026-10-24",
   "gpuwerk": "/docs/models/deepseek-v4-flash/",
   "spark": {
    "tps": "13 to 55",
    "kind": "reported",
    "detail": "one Spark needs a 2-bit or reduced build; range across engines"
   },
   "hf_check": {
    "repo": "deepseek-ai/DeepSeek-V4-Flash",
    "hf_total_b": 290.94,
    "agrees": true,
    "note": null
   },
   "variants": [],
   "eci": null,
   "shortlist": true
  },
  {
   "slug": "qwen3-235b-a22b",
   "name": "Qwen3-235B-A22B",
   "maker": "Qwen",
   "total_b": 235,
   "active_b": 22,
   "active_source": "card",
   "arch": "moe",
   "use": "Frontier-class open MoE",
   "source": "https://huggingface.co/Qwen/Qwen3-235B-A22B",
   "checked": "2026-09-07",
   "gpuwerk": "/docs/models/qwen3-235b-a22b/",
   "spark": null,
   "hf_check": {
    "repo": "Qwen/Qwen3-235B-A22B",
    "hf_total_b": 235.09,
    "agrees": true,
    "note": null
   },
   "variants": [
    {
     "label": "Qwen3 235B A22B",
     "quant": "Q4_K_M",
     "params_b": 235,
     "load_gb": 130,
     "min_ram_gb": 192,
     "kv_kb_per_token": null,
     "ollama": "ollama run qwen3:235b-a22b-q4_K_M",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Quality, Reasoning",
     "mf_id": null
    }
   ],
   "eci": null,
   "shortlist": true
  },
  {
   "slug": "qwen3-8-flash-next",
   "name": "Qwen3.8-Flash-Next",
   "maker": "Qwen",
   "total_b": 125,
   "active_b": 6,
   "active_source": "card",
   "arch": "moe",
   "use": "Large fast MoE for chat and agents",
   "source": "https://huggingface.co/Qwen/Qwen3.8-Flash-Next",
   "checked": "2026-10-11",
   "gpuwerk": "/docs/models/qwen3-8-flash-next/",
   "spark": null,
   "hf_check": {
    "repo": "Qwen/Qwen3.8-Flash-Next",
    "hf_total_b": 180.0,
    "agrees": false,
    "note": "Hugging Face total includes a 51B n-gram table and a 4B MTP head on top of the 125B model."
   },
   "variants": [
    {
     "label": "Qwen3.8-Flash-Next",
     "quant": "Q4_K_XL",
     "params_b": 125,
     "load_gb": 111.3,
     "min_ram_gb": 192,
     "kv_kb_per_token": null,
     "ollama": null,
     "runtimes": [
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Agentic coding, Reasoning, Multimodal",
     "mf_id": "qwen3.8-flash-next"
    },
    {
     "label": "Qwen3.8-Flash-Next (IQ4_XS)",
     "quant": "IQ4_XS",
     "params_b": 125,
     "load_gb": 93.7,
     "min_ram_gb": 128,
     "kv_kb_per_token": null,
     "ollama": null,
     "runtimes": [
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Agentic coding, Reasoning, Multimodal",
     "mf_id": "qwen3.8-flash-next"
    },
    {
     "label": "Qwen3.8-Flash-Next (Q3)",
     "quant": "Q3_K_XL",
     "params_b": 125,
     "load_gb": 90,
     "min_ram_gb": 128,
     "kv_kb_per_token": null,
     "ollama": null,
     "runtimes": [
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Agentic coding, Reasoning, Multimodal",
     "mf_id": "qwen3.8-flash-next"
    }
   ],
   "eci": null,
   "shortlist": true
  },
  {
   "slug": "nemotron-3-super-120b-a12b",
   "name": "Nemotron 3 Super 120B-A12B",
   "maker": "NVIDIA",
   "total_b": 120,
   "active_b": 12,
   "active_source": "card",
   "arch": "moe",
   "use": "Large MoE for agents and RAG",
   "source": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-NVFP4",
   "checked": "2026-09-07",
   "gpuwerk": "/docs/models/nemotron-3-super-120b/",
   "spark": {
    "tps": "22.7 to 23.7",
    "kind": "reported",
    "detail": "NVFP4 under stock vLLM, single user, per the vLLM project"
   },
   "hf_check": null,
   "variants": [
    {
     "label": "Nemotron 3 Super 120B-A12B",
     "quant": "Q4_K_M",
     "params_b": 120,
     "load_gb": 86.8,
     "min_ram_gb": 128,
     "kv_kb_per_token": 8,
     "ollama": "ollama run nemotron-3-super:120b",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Agentic workflows, Long context, Reasoning",
     "mf_id": "nemotron-3-super-120b-q4"
    },
    {
     "label": "Nemotron 3 Super 120B-A12B (IQ4_XS)",
     "quant": "IQ4_XS",
     "params_b": 120,
     "load_gb": 64.5,
     "min_ram_gb": 96,
     "kv_kb_per_token": 8,
     "ollama": null,
     "runtimes": [
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Agentic workflows, Long context, Reasoning",
     "mf_id": null
    }
   ],
   "eci": null,
   "shortlist": true
  },
  {
   "slug": "nemotron-3-5-super-vl",
   "name": "Nemotron 3.5 Super VL 120B-A12B",
   "maker": "NVIDIA",
   "total_b": 120,
   "active_b": 12,
   "active_source": "card",
   "arch": "moe",
   "use": "Vision-language MoE",
   "source": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-BF16",
   "checked": "2026-10-15",
   "gpuwerk": "/docs/models/nemotron-3-5-super/",
   "spark": null,
   "hf_check": {
    "repo": "nvidia/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-BF16",
    "hf_total_b": 124.45,
    "agrees": true,
    "note": null
   },
   "variants": [],
   "eci": null,
   "shortlist": true
  },
  {
   "slug": "gpt-oss-120b",
   "name": "gpt-oss-120b",
   "maker": "OpenAI",
   "total_b": 117,
   "active_b": 5.1,
   "active_source": "card",
   "arch": "moe",
   "use": "General chat and reasoning",
   "source": "https://huggingface.co/openai/gpt-oss-120b",
   "checked": "2026-09-07",
   "gpuwerk": "/docs/models/gpt-oss-120b/",
   "spark": {
    "tps": "60.5",
    "kind": "reported",
    "detail": "decode, one Spark, published measurement"
   },
   "hf_check": {
    "repo": "openai/gpt-oss-120b",
    "hf_total_b": 116.83,
    "agrees": true,
    "note": null
   },
   "variants": [
    {
     "label": "GPT-OSS 120B",
     "quant": "MXFP4",
     "params_b": 117,
     "load_gb": 65.4,
     "min_ram_gb": 96,
     "kv_kb_per_token": null,
     "ollama": "ollama run gpt-oss:120b",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Reasoning, Coding, Agents",
     "mf_id": "gpt-oss-120b"
    }
   ],
   "eci": null,
   "shortlist": true
  },
  {
   "slug": "llama-3-3-70b",
   "name": "Llama 3.3 70B",
   "maker": "Meta",
   "total_b": 70.6,
   "active_b": 70.6,
   "active_source": "card",
   "arch": "dense",
   "use": "Well-known dense 70B",
   "source": "https://huggingface.co/meta-llama/Llama-3.3-70B-Instruct",
   "checked": "2026-09-07",
   "gpuwerk": "/docs/models/llama-3-3-70b/",
   "spark": {
    "tps": "6.0",
    "kind": "measured",
    "detail": "AWQ 4-bit under vLLM, single user, measured by GPUwerk on 2 September 2026"
   },
   "hf_check": {
    "repo": "meta-llama/Llama-3.3-70B-Instruct",
    "hf_total_b": 70.55,
    "agrees": true,
    "note": null
   },
   "variants": [
    {
     "label": "Llama 3.3 70B Instruct",
     "quant": "Q4_K_M",
     "params_b": 70,
     "load_gb": 42,
     "min_ram_gb": 64,
     "kv_kb_per_token": null,
     "ollama": "ollama run llama3.3:70b-instruct-q4_K_M",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Quality, Coding",
     "mf_id": "llama3.3-70b-q4"
    },
    {
     "label": "Llama 3.3 70B Instruct (Q8)",
     "quant": "Q8_0",
     "params_b": 70,
     "load_gb": 75,
     "min_ram_gb": 96,
     "kv_kb_per_token": null,
     "ollama": "ollama run llama3.3:70b-instruct-q8_0",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Quality, Coding",
     "mf_id": "llama3.3-70b-q8"
    },
    {
     "label": "Llama 3.3 70B Instruct (Q6)",
     "quant": "Q6_K",
     "params_b": 70,
     "load_gb": 57.9,
     "min_ram_gb": 96,
     "kv_kb_per_token": null,
     "ollama": "ollama run llama3.3:70b-instruct-q6_K",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Quality, Coding",
     "mf_id": "llama3.3-70b-q6"
    }
   ],
   "eci": null,
   "shortlist": true
  },
  {
   "slug": "nemotron-super-49b",
   "name": "Llama 3.3 Nemotron Super 49B",
   "maker": "NVIDIA",
   "total_b": 49,
   "active_b": 49,
   "active_source": "card",
   "arch": "dense",
   "use": "Dense model for batch work",
   "source": "https://huggingface.co/nvidia/Llama-3_3-Nemotron-Super-49B-v1_5",
   "checked": "2026-09-07",
   "gpuwerk": "/docs/models/nemotron-super-49b/",
   "spark": null,
   "hf_check": {
    "repo": "nvidia/Llama-3_3-Nemotron-Super-49B-v1_5",
    "hf_total_b": 49.87,
    "agrees": true,
    "note": null
   },
   "variants": [],
   "eci": null,
   "shortlist": true
  },
  {
   "slug": "qwen3-6-35b-a3b",
   "name": "Qwen3.6-35B-A3B",
   "maker": "Qwen",
   "total_b": 35,
   "active_b": 3,
   "active_source": "card",
   "arch": "moe",
   "use": "Fast mid-size MoE, long context",
   "source": "https://huggingface.co/Qwen/Qwen3.6-35B-A3B",
   "checked": "2026-10-11",
   "gpuwerk": "/docs/models/qwen3-6-35b-a3b/",
   "spark": null,
   "hf_check": {
    "repo": "Qwen/Qwen3.6-35B-A3B",
    "hf_total_b": 35.95,
    "agrees": true,
    "note": null
   },
   "variants": [
    {
     "label": "Qwen3.6 35B-A3B",
     "quant": "Q4_K_M",
     "params_b": 35,
     "load_gb": 22,
     "min_ram_gb": 32,
     "kv_kb_per_token": 20,
     "ollama": "ollama run qwen3.6:35b-a3b",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Reasoning, Coding, Agents",
     "mf_id": "qwen3.6-35b-a3b-q4"
    },
    {
     "label": "Qwen3.6 35B-A3B (Q8)",
     "quant": "Q8_0",
     "params_b": 35,
     "load_gb": 38.7,
     "min_ram_gb": 64,
     "kv_kb_per_token": 20,
     "ollama": "ollama run qwen3.6:35b-a3b-q8_0",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Reasoning, Coding, Agents",
     "mf_id": "qwen3.6-35b-a3b-q8"
    }
   ],
   "eci": 143.81,
   "shortlist": true
  },
  {
   "slug": "gemma-4-31b",
   "name": "Gemma 4 31B",
   "maker": "Google",
   "total_b": 30.7,
   "active_b": 30.7,
   "active_source": "card",
   "arch": "dense",
   "use": "Dense multimodal model",
   "source": null,
   "checked": "2026-10-11",
   "gpuwerk": "/docs/models/gemma-4-26b-a4b/",
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Gemma 4 31B",
     "quant": "Q4_K_M",
     "params_b": 31,
     "load_gb": 20,
     "min_ram_gb": 32,
     "kv_kb_per_token": null,
     "ollama": "ollama run gemma4:31b",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Quality, Coding, Multimodal",
     "mf_id": "gemma4-31b-q4"
    },
    {
     "label": "Gemma 4 31B (Q8)",
     "quant": "Q8_0",
     "params_b": 31,
     "load_gb": 30.9,
     "min_ram_gb": 48,
     "kv_kb_per_token": null,
     "ollama": "ollama run gemma4:31b-it-q8_0",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Quality, Coding, Multimodal",
     "mf_id": "gemma4-31b-q8"
    }
   ],
   "eci": null,
   "shortlist": true
  },
  {
   "slug": "qwen3-coder-30b-a3b",
   "name": "Qwen3-Coder-30B-A3B",
   "maker": "Qwen",
   "total_b": 30.5,
   "active_b": 3.3,
   "active_source": "card",
   "arch": "moe",
   "use": "Coding assistant",
   "source": "https://huggingface.co/Qwen/Qwen3-Coder-30B-A3B-Instruct",
   "checked": "2026-09-07",
   "gpuwerk": "/docs/models/qwen3-coder-30b-a3b/",
   "spark": null,
   "hf_check": {
    "repo": "Qwen/Qwen3-Coder-30B-A3B-Instruct",
    "hf_total_b": 30.53,
    "agrees": true,
    "note": null
   },
   "variants": [
    {
     "label": "Qwen3 Coder 30B-A3B",
     "quant": "Q4_K_M",
     "params_b": 30.5,
     "load_gb": 18.6,
     "min_ram_gb": 32,
     "kv_kb_per_token": null,
     "ollama": "ollama run qwen3-coder:30b-a3b-q4_K_M",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Coding, Agentic coding, Long context",
     "mf_id": "qwen3-coder-30b-a3b-q4"
    },
    {
     "label": "Qwen3 Coder 30B-A3B (Q8)",
     "quant": "Q8_0",
     "params_b": 30.5,
     "load_gb": 32.5,
     "min_ram_gb": 48,
     "kv_kb_per_token": null,
     "ollama": "ollama run qwen3-coder:30b-a3b-q8_0",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Coding, Agentic coding, Long context",
     "mf_id": "qwen3-coder-30b-a3b-q8"
    }
   ],
   "eci": null,
   "shortlist": true
  },
  {
   "slug": "nemotron-3-nano-30b-a3b",
   "name": "Nemotron 3 Nano 30B-A3B",
   "maker": "NVIDIA",
   "total_b": 30,
   "active_b": 3,
   "active_source": "card",
   "arch": "moe",
   "use": "Fast default with room for context",
   "source": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-NVFP4",
   "checked": "2026-10-11",
   "gpuwerk": "/docs/models/",
   "spark": null,
   "hf_check": null,
   "variants": [],
   "eci": null,
   "shortlist": true
  },
  {
   "slug": "qwen3-8-27b",
   "name": "Qwen3.8-27B",
   "maker": "Qwen",
   "total_b": 27.3,
   "active_b": 27.3,
   "active_source": "card",
   "arch": "dense",
   "use": "Strong general dense model",
   "source": "https://huggingface.co/Qwen/Qwen3.8-27B",
   "checked": "2026-10-11",
   "gpuwerk": "/docs/models/qwen3-8-27b/",
   "spark": {
    "tps": "11.4 to 24.7",
    "kind": "reported",
    "detail": "range across quantizations and engines"
   },
   "hf_check": {
    "repo": "Qwen/Qwen3.8-27B",
    "hf_total_b": 27.78,
    "agrees": true,
    "note": null
   },
   "variants": [
    {
     "label": "Qwen3.8 27B",
     "quant": "Q4_K_M",
     "params_b": 27,
     "load_gb": 16.5,
     "min_ram_gb": 24,
     "kv_kb_per_token": 64,
     "ollama": "ollama run qwen3.8:27b",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Coding, Agent, Vision, Long context",
     "mf_id": "qwen3.8-27b-q4"
    },
    {
     "label": "Qwen3.8 27B (Q8)",
     "quant": "Q8_0",
     "params_b": 27,
     "load_gb": 27.1,
     "min_ram_gb": 48,
     "kv_kb_per_token": 64,
     "ollama": "ollama run qwen3.8:27b-q8_0",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Coding, Agent, Vision, Long context",
     "mf_id": "qwen3.8-27b-q8"
    }
   ],
   "eci": 151.8,
   "shortlist": true
  },
  {
   "slug": "gemma-4-26b-a4b",
   "name": "Gemma 4 26B-A4B",
   "maker": "Google",
   "total_b": 25.2,
   "active_b": 3.8,
   "active_source": "card",
   "arch": "moe",
   "use": "Fast multimodal MoE (text and image)",
   "source": "https://huggingface.co/google/gemma-4-26B-A4B",
   "checked": "2026-10-11",
   "gpuwerk": "/docs/models/gemma-4-26b-a4b/",
   "spark": {
    "tps": "52",
    "kind": "reported",
    "detail": "single user"
   },
   "hf_check": {
    "repo": "google/gemma-4-26B-A4B",
    "hf_total_b": 25.81,
    "agrees": true,
    "note": null
   },
   "variants": [
    {
     "label": "Gemma 4 26B-A4B",
     "quant": "Q4_K_M",
     "params_b": 26,
     "load_gb": 16,
     "min_ram_gb": 24,
     "kv_kb_per_token": null,
     "ollama": "ollama run gemma4:26b",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Chat, Coding, Multimodal",
     "mf_id": "gemma4-26b-a4b-q4"
    },
    {
     "label": "Gemma 4 26B-A4B (Q8)",
     "quant": "Q8_0",
     "params_b": 26,
     "load_gb": 28.1,
     "min_ram_gb": 48,
     "kv_kb_per_token": null,
     "ollama": "ollama run gemma4:26b-a4b-it-q8_0",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Chat, Coding, Multimodal",
     "mf_id": "gemma4-26b-a4b-q8"
    }
   ],
   "eci": 141.95,
   "shortlist": true
  },
  {
   "slug": "gpt-oss-20b",
   "name": "gpt-oss-20b",
   "maker": "OpenAI",
   "total_b": 21,
   "active_b": 3.6,
   "active_source": "card",
   "arch": "moe",
   "use": "General chat on modest hardware",
   "source": "https://huggingface.co/openai/gpt-oss-20b",
   "checked": "2026-09-07",
   "gpuwerk": "/docs/models/gpt-oss-20b/",
   "spark": null,
   "hf_check": {
    "repo": "openai/gpt-oss-20b",
    "hf_total_b": 20.91,
    "agrees": true,
    "note": null
   },
   "variants": [
    {
     "label": "GPT-OSS 20B",
     "quant": "MXFP4",
     "params_b": 21,
     "load_gb": 13.8,
     "min_ram_gb": 24,
     "kv_kb_per_token": null,
     "ollama": "ollama run gpt-oss:20b",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Chat, Coding, Reasoning",
     "mf_id": "gpt-oss-20b"
    }
   ],
   "eci": 137.75,
   "shortlist": true
  },
  {
   "slug": "nemotron-nano-9b-v2",
   "name": "Nemotron Nano 9B v2",
   "maker": "NVIDIA",
   "total_b": 9,
   "active_b": 9,
   "active_source": "card",
   "arch": "dense",
   "use": "Small fast model, draft model",
   "source": "https://huggingface.co/nvidia/NVIDIA-Nemotron-Nano-9B-v2-NVFP4",
   "checked": "2026-10-11",
   "gpuwerk": "/docs/models/",
   "spark": null,
   "hf_check": null,
   "variants": [],
   "eci": null,
   "shortlist": true
  },
  {
   "slug": "deepseek-r1-671b",
   "name": "DeepSeek-R1 671B",
   "maker": "DeepSeek",
   "total_b": 671,
   "active_b": 37,
   "active_source": "card",
   "arch": "moe",
   "use": "Reasoning, Coding",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "DeepSeek-R1 671B",
     "quant": "Q4_K_M",
     "params_b": 671,
     "load_gb": 380,
     "min_ram_gb": 512,
     "kv_kb_per_token": null,
     "ollama": "ollama run deepseek-r1:671b-q4_K_M",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Reasoning, Coding",
     "mf_id": "deepseek-r1-671b-q4"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "ling-3.1-flash-1019",
   "name": "Ling-3.1-flash-1019",
   "maker": "Ling",
   "total_b": 560,
   "active_b": null,
   "active_source": null,
   "arch": null,
   "use": "Reasoning, Agents, Long context",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Ling-3.1-flash-1019",
     "quant": "Q4_K_M",
     "params_b": 560,
     "load_gb": 358,
     "min_ram_gb": 512,
     "kv_kb_per_token": null,
     "ollama": null,
     "runtimes": [
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Reasoning, Agents, Long context",
     "mf_id": "ling-3.1-flash-1019"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "llama-3.1-405b-instruct",
   "name": "Llama 3.1 405B Instruct",
   "maker": "Llama",
   "total_b": 405,
   "active_b": 405,
   "active_source": "dense",
   "arch": "dense",
   "use": "Quality, Reasoning, Coding",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Llama 3.1 405B Instruct",
     "quant": "Q4_K_M",
     "params_b": 405,
     "load_gb": 243,
     "min_ram_gb": 320,
     "kv_kb_per_token": null,
     "ollama": "ollama run llama3.1:405b-instruct-q4_K_M",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Quality, Reasoning, Coding",
     "mf_id": "llama-3.1-405b-q4"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "llama-4-maverick",
   "name": "Llama 4 Maverick",
   "maker": "Llama",
   "total_b": 400,
   "active_b": 17,
   "active_source": "card",
   "arch": "moe",
   "use": "Frontier quality, Long context",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Llama 4 Maverick",
     "quant": "Q4_K_M",
     "params_b": 400,
     "load_gb": 245,
     "min_ram_gb": 320,
     "kv_kb_per_token": null,
     "ollama": "ollama run llama4:maverick",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Frontier quality, Long context",
     "mf_id": "llama4-maverick-q4"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "ornith-1.5-122b",
   "name": "Ornith 1.5 122B",
   "maker": "Ornith",
   "total_b": 125.09,
   "active_b": null,
   "active_source": null,
   "arch": null,
   "use": "Agentic coding",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Ornith 1.5 122B",
     "quant": "Q4_K_M",
     "params_b": 125.09,
     "load_gb": 75.8,
     "min_ram_gb": 128,
     "kv_kb_per_token": 24,
     "ollama": "ollama run ornith-1.5:122b",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Agentic coding",
     "mf_id": "ornith-1.5-122b-q4"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "ling-3.0-flash",
   "name": "Ling-3.0-flash",
   "maker": "Ling",
   "total_b": 124,
   "active_b": null,
   "active_source": null,
   "arch": null,
   "use": "Agentic coding, Long context, Reasoning",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Ling-3.0-flash",
     "quant": "Q4_K_M",
     "params_b": 124,
     "load_gb": 77.8,
     "min_ram_gb": 128,
     "kv_kb_per_token": null,
     "ollama": null,
     "runtimes": [
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Agentic coding, Long context, Reasoning",
     "mf_id": "ling-3.0-flash"
    },
    {
     "label": "Ling-3.0-flash (IQ4_XS)",
     "quant": "IQ4_XS",
     "params_b": 124,
     "load_gb": 68.7,
     "min_ram_gb": 96,
     "kv_kb_per_token": null,
     "ollama": null,
     "runtimes": [
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Agentic coding, Long context, Reasoning",
     "mf_id": "ling-3.0-flash"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "devstral-2-123b",
   "name": "Devstral 2 123B",
   "maker": "Mistral",
   "total_b": 123,
   "active_b": null,
   "active_source": null,
   "arch": null,
   "use": "Agentic coding, Software engineering",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Devstral 2 123B",
     "quant": "Q4_K_M",
     "params_b": 123,
     "load_gb": 74.9,
     "min_ram_gb": 96,
     "kv_kb_per_token": null,
     "ollama": "ollama run devstral-2:123b",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Agentic coding, Software engineering",
     "mf_id": "devstral-2-123b-q4"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "qwen3.5-122b-a10b-instruct",
   "name": "Qwen3.5 122B-A10B Instruct",
   "maker": "Qwen",
   "total_b": 122,
   "active_b": 10.0,
   "active_source": "name",
   "arch": "moe",
   "use": "Frontier-level reasoning, Complex tasks",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Qwen3.5 122B-A10B Instruct",
     "quant": "Q4_K_M",
     "params_b": 122,
     "load_gb": 72,
     "min_ram_gb": 96,
     "kv_kb_per_token": 24,
     "ollama": "ollama run qwen3.5:122b-a10b",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Frontier-level reasoning, Complex tasks",
     "mf_id": "qwen3.5-122b-a10b-q4"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "laguna-s-2.1",
   "name": "Laguna S 2.1",
   "maker": "Laguna",
   "total_b": 118,
   "active_b": null,
   "active_source": null,
   "arch": null,
   "use": "Agentic coding, Long-horizon tasks",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Laguna S 2.1",
     "quant": "Q4_K_M",
     "params_b": 118,
     "load_gb": 96,
     "min_ram_gb": 128,
     "kv_kb_per_token": null,
     "ollama": "ollama run laguna-s-2.1:q4_K_M",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Agentic coding, Long-horizon tasks",
     "mf_id": "laguna-s-2.1-q4"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "llama-4-scout",
   "name": "Llama 4 Scout",
   "maker": "Llama",
   "total_b": 109,
   "active_b": 17,
   "active_source": "card",
   "arch": "moe",
   "use": "Long context, Quality, Multimodal",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Llama 4 Scout",
     "quant": "Q4_K_M",
     "params_b": 109,
     "load_gb": 67,
     "min_ram_gb": 96,
     "kv_kb_per_token": null,
     "ollama": "ollama run llama4:scout",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Long context, Quality, Multimodal",
     "mf_id": "llama4-scout-q4"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "qwen3-next-80b-a3b",
   "name": "Qwen3-Next 80B-A3B",
   "maker": "Qwen",
   "total_b": 80,
   "active_b": 3.0,
   "active_source": "name",
   "arch": "moe",
   "use": "Chat, Coding, Long Context",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Qwen3-Next 80B-A3B",
     "quant": "Q4_K_M",
     "params_b": 80,
     "load_gb": 50.4,
     "min_ram_gb": 72,
     "kv_kb_per_token": 24,
     "ollama": "ollama run qwen3-next:80b",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Chat, Coding, Long Context",
     "mf_id": "qwen3-next-80b-a3b"
    },
    {
     "label": "Qwen3-Next 80B-A3B (Q8)",
     "quant": "Q8_0",
     "params_b": 80,
     "load_gb": 84.8,
     "min_ram_gb": 128,
     "kv_kb_per_token": 24,
     "ollama": "ollama run qwen3-next:80b-a3b-instruct-q8_0",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Chat, Coding, Long Context",
     "mf_id": "qwen3-next-80b-a3b"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "kolibri-1",
   "name": "Kolibri 1",
   "maker": "Kolibri",
   "total_b": 78.1,
   "active_b": null,
   "active_source": null,
   "arch": null,
   "use": "Reasoning, German, Long context",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Kolibri 1",
     "quant": "Q4_K_M",
     "params_b": 78.1,
     "load_gb": 49.8,
     "min_ram_gb": 96,
     "kv_kb_per_token": 20,
     "ollama": null,
     "runtimes": [
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Reasoning, German, Long context",
     "mf_id": "kolibri-1-q4"
    },
    {
     "label": "Kolibri 1 (IQ4_XS)",
     "quant": "IQ4_XS",
     "params_b": 78.1,
     "load_gb": 43.6,
     "min_ram_gb": 64,
     "kv_kb_per_token": 20,
     "ollama": null,
     "runtimes": [
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Reasoning, German, Long context",
     "mf_id": "kolibri-1-q4xs"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "llama-3.1-70b-instruct",
   "name": "Llama 3.1 70B Instruct",
   "maker": "Llama",
   "total_b": 70,
   "active_b": 70,
   "active_source": "dense",
   "arch": "dense",
   "use": "Quality, Coding",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Llama 3.1 70B Instruct",
     "quant": "Q4_K_M",
     "params_b": 70,
     "load_gb": 42,
     "min_ram_gb": 64,
     "kv_kb_per_token": null,
     "ollama": "ollama run llama3.1:70b-instruct-q4_K_M",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Quality, Coding",
     "mf_id": "llama3.1-70b-q4"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "deepseek-r1-distill-llama-70b",
   "name": "DeepSeek-R1 Distill Llama 70B",
   "maker": "DeepSeek",
   "total_b": 70,
   "active_b": 70,
   "active_source": "dense",
   "arch": "dense",
   "use": "Reasoning, Quality",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "DeepSeek-R1 Distill Llama 70B",
     "quant": "Q4_K_M",
     "params_b": 70,
     "load_gb": 42,
     "min_ram_gb": 64,
     "kv_kb_per_token": null,
     "ollama": "ollama run deepseek-r1:70b",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Reasoning, Quality",
     "mf_id": "deepseek-r1-distill-llama-70b-q4"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "mixtral-8x7b-instruct",
   "name": "Mixtral 8x7B Instruct",
   "maker": "Mistral",
   "total_b": 46.7,
   "active_b": 12.9,
   "active_source": "card",
   "arch": "moe",
   "use": "Coding, Quality",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Mixtral 8x7B Instruct",
     "quant": "Q4_K_M",
     "params_b": 46.7,
     "load_gb": 30,
     "min_ram_gb": 48,
     "kv_kb_per_token": null,
     "ollama": "ollama run mixtral:8x7b",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Coding, Quality",
     "mf_id": "mixtral-8x7b-q4"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "flint-40b-a2.5b",
   "name": "Flint 40B-A2.5B",
   "maker": "Flint",
   "total_b": 39.8,
   "active_b": 2.5,
   "active_source": "name",
   "arch": "moe",
   "use": "Agentic coding, Long context",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Flint 40B-A2.5B",
     "quant": "NVFP4",
     "params_b": 39.8,
     "load_gb": 22.9,
     "min_ram_gb": 36,
     "kv_kb_per_token": 28,
     "ollama": "ollama run flint:40b",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Agentic coding, Long context",
     "mf_id": "flint-40b"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "sage-a1.5",
   "name": "Sage-A1.5",
   "maker": "Sage",
   "total_b": 35.95,
   "active_b": null,
   "active_source": null,
   "arch": null,
   "use": "Scientific research agents, Long-horizon tasks",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Sage-A1.5",
     "quant": "Q4_K_M",
     "params_b": 35.95,
     "load_gb": 21.7,
     "min_ram_gb": 32,
     "kv_kb_per_token": 20,
     "ollama": null,
     "runtimes": [
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Scientific research agents, Long-horizon tasks",
     "mf_id": "sage-a1.5-q4"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "qwen3.5-35b-a3b-instruct",
   "name": "Qwen3.5 35B-A3B Instruct",
   "maker": "Qwen",
   "total_b": 35,
   "active_b": 3.0,
   "active_source": "name",
   "arch": "moe",
   "use": "Reasoning, Coding, Agent scenarios",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Qwen3.5 35B-A3B Instruct",
     "quant": "Q4_K_M",
     "params_b": 35,
     "load_gb": 20,
     "min_ram_gb": 32,
     "kv_kb_per_token": 20,
     "ollama": "ollama run qwen3.5:35b-a3b",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Reasoning, Coding, Agent scenarios",
     "mf_id": "qwen3.5-35b-a3b-q4"
    },
    {
     "label": "Qwen3.5 35B-A3B Instruct (Q8)",
     "quant": "Q8_0",
     "params_b": 35,
     "load_gb": 38.7,
     "min_ram_gb": 64,
     "kv_kb_per_token": 20,
     "ollama": "ollama run qwen3.5:35b-a3b-q8_0",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Reasoning, Coding, Agent scenarios",
     "mf_id": "qwen3.5-35b-a3b-q8"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "ornith-1.0-35b",
   "name": "Ornith 1.0 35B",
   "maker": "Ornith",
   "total_b": 35,
   "active_b": null,
   "active_source": null,
   "arch": null,
   "use": "Agentic coding",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Ornith 1.0 35B",
     "quant": "Q4_K_M",
     "params_b": 35,
     "load_gb": 21.2,
     "min_ram_gb": 32,
     "kv_kb_per_token": null,
     "ollama": "ollama run ornith:35b",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Agentic coding",
     "mf_id": "ornith-35b-q4"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "nanbeige4.3-33b",
   "name": "Nanbeige4.3 33B",
   "maker": "Nanbeige",
   "total_b": 33.1,
   "active_b": null,
   "active_source": null,
   "arch": null,
   "use": "Agentic coding, Agents, Long context",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Nanbeige4.3 33B",
     "quant": "Q4_K_M",
     "params_b": 33.1,
     "load_gb": 20.8,
     "min_ram_gb": 32,
     "kv_kb_per_token": null,
     "ollama": null,
     "runtimes": [
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Agentic coding, Agents, Long context",
     "mf_id": "nanbeige4.3-33b-q4"
    },
    {
     "label": "Nanbeige4.3 33B (Q8)",
     "quant": "Q8_0",
     "params_b": 33.1,
     "load_gb": 35.2,
     "min_ram_gb": 64,
     "kv_kb_per_token": null,
     "ollama": null,
     "runtimes": [
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Agentic coding, Agents, Long context",
     "mf_id": "nanbeige4.3-33b-q8"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "poolside-laguna-xs.2",
   "name": "Poolside Laguna XS.2",
   "maker": "Laguna",
   "total_b": 33,
   "active_b": null,
   "active_source": null,
   "arch": null,
   "use": "Agentic coding, long-horizon software engineering",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Poolside Laguna XS.2",
     "quant": "Q4_K_M",
     "params_b": 33,
     "load_gb": 23,
     "min_ram_gb": 36,
     "kv_kb_per_token": null,
     "ollama": "ollama run laguna-xs.2:q4_K_M",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Agentic coding, long-horizon software engineering",
     "mf_id": null
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "laguna-xs-2.1",
   "name": "Laguna XS 2.1",
   "maker": "Laguna",
   "total_b": 33,
   "active_b": null,
   "active_source": null,
   "arch": null,
   "use": "Agentic coding, Long-horizon tasks",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Laguna XS 2.1",
     "quant": "Q4_K_M",
     "params_b": 33,
     "load_gb": 20.3,
     "min_ram_gb": 32,
     "kv_kb_per_token": null,
     "ollama": "ollama run laguna-xs-2.1:q4_K_M",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Agentic coding, Long-horizon tasks",
     "mf_id": "laguna-xs-2.1-q4"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "qwen3-32b",
   "name": "Qwen3 32B",
   "maker": "Qwen",
   "total_b": 32,
   "active_b": 32,
   "active_source": "dense",
   "arch": "dense",
   "use": "Quality, Coding",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Qwen3 32B",
     "quant": "Q4_K_M",
     "params_b": 32,
     "load_gb": 20.2,
     "min_ram_gb": 32,
     "kv_kb_per_token": null,
     "ollama": "ollama run qwen3:32b-q4_K_M",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Quality, Coding",
     "mf_id": "qwen3-32b-q4"
    },
    {
     "label": "Qwen3 32B (Q8)",
     "quant": "Q8_0",
     "params_b": 32,
     "load_gb": 35.1,
     "min_ram_gb": 64,
     "kv_kb_per_token": null,
     "ollama": "ollama run qwen3:32b-q8_0",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Quality, Coding",
     "mf_id": "qwen3-32b-q8"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "deepseek-r1-distill-qwen-32b",
   "name": "DeepSeek-R1 Distill Qwen 32B",
   "maker": "DeepSeek",
   "total_b": 32,
   "active_b": 32,
   "active_source": "dense",
   "arch": "dense",
   "use": "Reasoning, Quality",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "DeepSeek-R1 Distill Qwen 32B",
     "quant": "Q4_K_M",
     "params_b": 32,
     "load_gb": 19.9,
     "min_ram_gb": 32,
     "kv_kb_per_token": null,
     "ollama": "ollama run deepseek-r1:32b-qwen-distill-q4_K_M",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Reasoning, Quality",
     "mf_id": "deepseek-r1-distill-qwen-32b-q4"
    },
    {
     "label": "DeepSeek-R1 Distill Qwen 32B (Q8)",
     "quant": "Q8_0",
     "params_b": 32,
     "load_gb": 34.8,
     "min_ram_gb": 64,
     "kv_kb_per_token": null,
     "ollama": "ollama run deepseek-r1:32b-qwen-distill-q8_0",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Reasoning, Quality",
     "mf_id": "deepseek-r1-distill-qwen-32b-q8"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "olmo-3.1-32b-instruct",
   "name": "Olmo 3.1 32B Instruct",
   "maker": "Olmo",
   "total_b": 32,
   "active_b": null,
   "active_source": null,
   "arch": null,
   "use": "General chat, Research, Fully open",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Olmo 3.1 32B Instruct",
     "quant": "Q4_K_M",
     "params_b": 32,
     "load_gb": 19.5,
     "min_ram_gb": 32,
     "kv_kb_per_token": null,
     "ollama": "ollama run olmo-3.1:32b",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "General chat, Research, Fully open",
     "mf_id": "olmo-3.1-32b-q4"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "qwen3-30b",
   "name": "Qwen3 30B",
   "maker": "Qwen",
   "total_b": 30,
   "active_b": 30,
   "active_source": "dense",
   "arch": "dense",
   "use": "Quality, Coding",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Qwen3 30B",
     "quant": "Q4_K_M",
     "params_b": 30,
     "load_gb": 22,
     "min_ram_gb": 32,
     "kv_kb_per_token": null,
     "ollama": "ollama run qwen3:30b-a3b-q4_K_M",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Quality, Coding",
     "mf_id": "qwen3-30b-q4"
    },
    {
     "label": "Qwen3 30B (Q8)",
     "quant": "Q8_0",
     "params_b": 30,
     "load_gb": 30.3,
     "min_ram_gb": 48,
     "kv_kb_per_token": null,
     "ollama": "ollama run qwen3:30b-a3b-q8_0",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Quality, Coding",
     "mf_id": "qwen3-30b-a3b-q8"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "nvidia-nemotron-cascade-2-30b-a3b",
   "name": "NVIDIA Nemotron Cascade 2 30B-A3B",
   "maker": "Nemotron",
   "total_b": 30,
   "active_b": 3.0,
   "active_source": "name",
   "arch": "moe",
   "use": "Reasoning, Math, Agentic tasks",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "NVIDIA Nemotron Cascade 2 30B-A3B",
     "quant": "Q6_K",
     "params_b": 30,
     "load_gb": 24,
     "min_ram_gb": 36,
     "kv_kb_per_token": 6,
     "ollama": "ollama run nemotron-cascade-2:30b",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Reasoning, Math, Agentic tasks",
     "mf_id": null
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "cohere-north-mini-code",
   "name": "Cohere North Mini Code",
   "maker": "North",
   "total_b": 30,
   "active_b": null,
   "active_source": null,
   "arch": null,
   "use": "Agentic coding, repo-scale software engineering",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Cohere North Mini Code",
     "quant": "Q4_K_M",
     "params_b": 30,
     "load_gb": 19,
     "min_ram_gb": 32,
     "kv_kb_per_token": null,
     "ollama": "ollama run north-mini-code-1.0:q4_K_M",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Agentic coding, repo-scale software engineering",
     "mf_id": null
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "nemotron-3.5-lightning-30b-a3b",
   "name": "Nemotron 3.5 Lightning 30B-A3B",
   "maker": "Nemotron",
   "total_b": 30,
   "active_b": 3.0,
   "active_source": "name",
   "arch": "moe",
   "use": "Agentic, Coding, Long context",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Nemotron 3.5 Lightning 30B-A3B",
     "quant": "Q4_K_M",
     "params_b": 30,
     "load_gb": 25.4,
     "min_ram_gb": 36,
     "kv_kb_per_token": 6,
     "ollama": "ollama run nemotron-3.5-lightning:30b",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Agentic, Coding, Long context",
     "mf_id": "nemotron-3.5-lightning-30b-q4"
    },
    {
     "label": "Nemotron 3.5 Lightning 30B-A3B (IQ4_XS)",
     "quant": "IQ4_XS",
     "params_b": 30,
     "load_gb": 18.9,
     "min_ram_gb": 32,
     "kv_kb_per_token": 6,
     "ollama": null,
     "runtimes": [
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Agentic, Coding, Long context",
     "mf_id": null
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "glm-4.7-flash-30b-a3b",
   "name": "GLM-4.7-Flash 30B-A3B",
   "maker": "Zhipu",
   "total_b": 30,
   "active_b": 3.0,
   "active_source": "name",
   "arch": "moe",
   "use": "Coding, Agentic tasks, Reasoning",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "GLM-4.7-Flash 30B-A3B",
     "quant": "Q4_K_M",
     "params_b": 30,
     "load_gb": 19,
     "min_ram_gb": 32,
     "kv_kb_per_token": null,
     "ollama": "ollama run glm-4.7-flash",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Coding, Agentic tasks, Reasoning",
     "mf_id": "glm-4.7-flash-q4"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "granite-switch-4.1-30b",
   "name": "Granite Switch 4.1 30B",
   "maker": "Granite",
   "total_b": 30,
   "active_b": null,
   "active_source": null,
   "arch": null,
   "use": "Chat, Decisions, Safety checks",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Granite Switch 4.1 30B",
     "quant": "Q4_K_M",
     "params_b": 30,
     "load_gb": 21,
     "min_ram_gb": 32,
     "kv_kb_per_token": null,
     "ollama": null,
     "runtimes": [
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Chat, Decisions, Safety checks",
     "mf_id": "granite-switch-4.1-30b-q4"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "muse-glimmer-30b",
   "name": "Muse Glimmer 30B",
   "maker": "Muse",
   "total_b": 29.8,
   "active_b": null,
   "active_source": null,
   "arch": null,
   "use": "Chat",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Muse Glimmer 30B",
     "quant": "Q4_K_M",
     "params_b": 29.8,
     "load_gb": 16.9,
     "min_ram_gb": 32,
     "kv_kb_per_token": null,
     "ollama": "ollama run muse-glimmer:30b",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Chat",
     "mf_id": "muse-glimmer-30b-q4"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "granite-4.2-30b",
   "name": "Granite 4.2 30B",
   "maker": "Granite",
   "total_b": 29.3,
   "active_b": null,
   "active_source": null,
   "arch": null,
   "use": "Coding, Chat",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Granite 4.2 30B",
     "quant": "Q4_K_M",
     "params_b": 29.3,
     "load_gb": 16.5,
     "min_ram_gb": 24,
     "kv_kb_per_token": null,
     "ollama": "ollama run granite4.2:30b",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Coding, Chat",
     "mf_id": "granite4.2-30b-q4"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "arex-2-27b",
   "name": "AREX-2 27B",
   "maker": "AREX",
   "total_b": 27.4,
   "active_b": null,
   "active_source": null,
   "arch": null,
   "use": "Agents, Deep research, Coding",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "AREX-2 27B",
     "quant": "Q4_K_M",
     "params_b": 27.4,
     "load_gb": 17.2,
     "min_ram_gb": 32,
     "kv_kb_per_token": 64,
     "ollama": null,
     "runtimes": [
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Agents, Deep research, Coding",
     "mf_id": "arex-2-27b-q4"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "gemma-2-27b-instruct",
   "name": "Gemma 2 27B Instruct",
   "maker": "Gemma",
   "total_b": 27,
   "active_b": 27,
   "active_source": "dense",
   "arch": "dense",
   "use": "Quality, Coding",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Gemma 2 27B Instruct",
     "quant": "Q4_K_M",
     "params_b": 27,
     "load_gb": 21,
     "min_ram_gb": 32,
     "kv_kb_per_token": null,
     "ollama": "ollama run gemma2:27b-instruct-q4_K_M",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Quality, Coding",
     "mf_id": "gemma2-27b-q4"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "gemma-3-27b-instruct",
   "name": "Gemma 3 27B Instruct",
   "maker": "Gemma",
   "total_b": 27,
   "active_b": 27,
   "active_source": "dense",
   "arch": "dense",
   "use": "Quality, Coding",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Gemma 3 27B Instruct",
     "quant": "Q4_K_M",
     "params_b": 27,
     "load_gb": 21,
     "min_ram_gb": 32,
     "kv_kb_per_token": null,
     "ollama": "ollama run gemma3:27b",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Quality, Coding",
     "mf_id": "gemma3-27b-q4"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "qwen3.5-27b-instruct",
   "name": "Qwen3.5 27B Instruct",
   "maker": "Qwen",
   "total_b": 27,
   "active_b": 27,
   "active_source": "dense",
   "arch": "dense",
   "use": "Chat, Coding, Complex reasoning",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Qwen3.5 27B Instruct",
     "quant": "Q4_K_M",
     "params_b": 27,
     "load_gb": 16,
     "min_ram_gb": 24,
     "kv_kb_per_token": 64,
     "ollama": "ollama run qwen3.5:27b",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Chat, Coding, Complex reasoning",
     "mf_id": "qwen3.5-27b-q4"
    },
    {
     "label": "Qwen3.5 27B Instruct (Q8)",
     "quant": "Q8_0",
     "params_b": 27,
     "load_gb": 27.1,
     "min_ram_gb": 48,
     "kv_kb_per_token": 64,
     "ollama": "ollama run qwen3.5:27b-q8_0",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Chat, Coding, Complex reasoning",
     "mf_id": "qwen3.5-27b-q8"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "qwen3.6-27b",
   "name": "Qwen3.6 27B",
   "maker": "Qwen",
   "total_b": 27,
   "active_b": 27,
   "active_source": "dense",
   "arch": "dense",
   "use": "Coding, Quality, Long context",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Qwen3.6 27B",
     "quant": "Q4_K_M",
     "params_b": 27,
     "load_gb": 18,
     "min_ram_gb": 32,
     "kv_kb_per_token": 64,
     "ollama": "ollama run qwen3.6:27b",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Coding, Quality, Long context",
     "mf_id": "qwen3.6-27b-q4"
    },
    {
     "label": "Qwen3.6 27B (Q8)",
     "quant": "Q8_0",
     "params_b": 27,
     "load_gb": 30,
     "min_ram_gb": 48,
     "kv_kb_per_token": 64,
     "ollama": "ollama run qwen3.6:27b-q8_0",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Coding, Quality, Long context",
     "mf_id": "qwen3.6-27b-q8"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "jebadiah-27b",
   "name": "Jebadiah 27B",
   "maker": "Jebadiah",
   "total_b": 27,
   "active_b": null,
   "active_source": null,
   "arch": null,
   "use": "Decisions",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Jebadiah 27B",
     "quant": "Q4_K_M",
     "params_b": 27,
     "load_gb": 16.8,
     "min_ram_gb": 32,
     "kv_kb_per_token": 64,
     "ollama": null,
     "runtimes": [
      "llama.cpp"
     ],
     "best_for": "Decisions",
     "mf_id": "jebadiah-27b-q4"
    },
    {
     "label": "Jebadiah 27B (Q5)",
     "quant": "Q5_K_M",
     "params_b": 27,
     "load_gb": 19.5,
     "min_ram_gb": 32,
     "kv_kb_per_token": 64,
     "ollama": null,
     "runtimes": [
      "llama.cpp"
     ],
     "best_for": "Decisions",
     "mf_id": "jebadiah-27b-q5"
    },
    {
     "label": "Jebadiah 27B (Q8)",
     "quant": "Q8_0",
     "params_b": 27,
     "load_gb": 29,
     "min_ram_gb": 48,
     "kv_kb_per_token": 64,
     "ollama": null,
     "runtimes": [
      "llama.cpp"
     ],
     "best_for": "Decisions",
     "mf_id": "jebadiah-27b-q8"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "mistral-small-3.1",
   "name": "Mistral Small 3.1",
   "maker": "Mistral",
   "total_b": 24,
   "active_b": 24,
   "active_source": "dense",
   "arch": "dense",
   "use": "Chat, Coding",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Mistral Small 3.1",
     "quant": "Q4_K_M",
     "params_b": 24,
     "load_gb": 15,
     "min_ram_gb": 24,
     "kv_kb_per_token": null,
     "ollama": "ollama run mistral-small3.1:24b",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Chat, Coding",
     "mf_id": null
    },
    {
     "label": "Mistral Small 3.1 (Q8)",
     "quant": "Q8_0",
     "params_b": 24,
     "load_gb": 23.3,
     "min_ram_gb": 36,
     "kv_kb_per_token": null,
     "ollama": "ollama run mistral-small3.1:24b-instruct-2503-q8_0",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Chat, Coding",
     "mf_id": "mistral-small-3.1-24b-q8"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "lfm2-24b-a2b-instruct",
   "name": "LFM2 24B-A2B Instruct",
   "maker": "LFM2",
   "total_b": 24,
   "active_b": 2.0,
   "active_source": "name",
   "arch": "moe",
   "use": "Local AI agents, privacy-first tool calling, MCP workflows",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "LFM2 24B-A2B Instruct",
     "quant": "Q4_K_M",
     "params_b": 24,
     "load_gb": 14,
     "min_ram_gb": 24,
     "kv_kb_per_token": null,
     "ollama": "ollama run lfm2:24b-a2b",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Local AI agents, privacy-first tool calling, MCP workflows",
     "mf_id": "lfm2-24b-a2b-q4"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "devstral-small-2-24b",
   "name": "Devstral Small 2 24B",
   "maker": "Mistral",
   "total_b": 24,
   "active_b": null,
   "active_source": null,
   "arch": null,
   "use": "Agentic coding, Software engineering",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Devstral Small 2 24B",
     "quant": "Q4_K_M",
     "params_b": 24,
     "load_gb": 15.2,
     "min_ram_gb": 24,
     "kv_kb_per_token": null,
     "ollama": "ollama run devstral-small-2:24b",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Agentic coding, Software engineering",
     "mf_id": "devstral-small-2-24b-q4"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "mistral-small-22b",
   "name": "Mistral Small 22B",
   "maker": "Mistral",
   "total_b": 22,
   "active_b": 22,
   "active_source": "dense",
   "arch": "dense",
   "use": "Coding, Quality",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Mistral Small 22B",
     "quant": "Q4_K_M",
     "params_b": 22,
     "load_gb": 17,
     "min_ram_gb": 32,
     "kv_kb_per_token": null,
     "ollama": "ollama run mistral-small:22b",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Coding, Quality",
     "mf_id": "mistral-small-22b-q4"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "qwen2.5-14b-instruct",
   "name": "Qwen2.5 14B Instruct",
   "maker": "Qwen",
   "total_b": 14,
   "active_b": 14,
   "active_source": "dense",
   "arch": "dense",
   "use": "Coding, Chat",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Qwen2.5 14B Instruct",
     "quant": "Q4_K_M",
     "params_b": 14,
     "load_gb": 11,
     "min_ram_gb": 16,
     "kv_kb_per_token": null,
     "ollama": "ollama run qwen2.5:14b-instruct-q4_K_M",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Coding, Chat",
     "mf_id": "qwen2.5-14b-q4"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "phi-3-medium-14b",
   "name": "Phi-3 Medium 14B",
   "maker": "Phi",
   "total_b": 14,
   "active_b": 14,
   "active_source": "dense",
   "arch": "dense",
   "use": "Coding, Quality",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Phi-3 Medium 14B",
     "quant": "Q4_K_M",
     "params_b": 14,
     "load_gb": 11,
     "min_ram_gb": 16,
     "kv_kb_per_token": null,
     "ollama": "ollama run phi3:14b-medium-4k-instruct-q4_K_M",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Coding, Quality",
     "mf_id": null
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "phi-4-14b",
   "name": "Phi-4 14B",
   "maker": "Phi",
   "total_b": 14,
   "active_b": 14,
   "active_source": "dense",
   "arch": "dense",
   "use": "Coding, Quality",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Phi-4 14B",
     "quant": "Q4_K_M",
     "params_b": 14,
     "load_gb": 11.5,
     "min_ram_gb": 24,
     "kv_kb_per_token": null,
     "ollama": "ollama run phi4:14b-q4_K_M",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Coding, Quality",
     "mf_id": "phi4-14b-q4"
    },
    {
     "label": "Phi-4 14B (Q8)",
     "quant": "Q8_0",
     "params_b": 14,
     "load_gb": 14.5,
     "min_ram_gb": 24,
     "kv_kb_per_token": null,
     "ollama": "ollama run phi4:14b-q8_0",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Coding, Quality",
     "mf_id": "phi4-14b-q8"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "qwen2.5-coder-14b",
   "name": "Qwen2.5 Coder 14B",
   "maker": "Qwen",
   "total_b": 14,
   "active_b": 14,
   "active_source": "dense",
   "arch": "dense",
   "use": "Coding",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Qwen2.5 Coder 14B",
     "quant": "Q4_K_M",
     "params_b": 14,
     "load_gb": 11,
     "min_ram_gb": 16,
     "kv_kb_per_token": null,
     "ollama": "ollama run qwen2.5-coder:14b",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Coding",
     "mf_id": "qwen2.5-coder-14b-q4"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "qwen3-14b",
   "name": "Qwen3 14B",
   "maker": "Qwen",
   "total_b": 14,
   "active_b": 14,
   "active_source": "dense",
   "arch": "dense",
   "use": "Coding, Quality",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Qwen3 14B",
     "quant": "Q4_K_M",
     "params_b": 14,
     "load_gb": 11,
     "min_ram_gb": 16,
     "kv_kb_per_token": null,
     "ollama": "ollama run qwen3:14b-q4_K_M",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Coding, Quality",
     "mf_id": "qwen3-14b-q4"
    },
    {
     "label": "Qwen3 14B (Q8)",
     "quant": "Q8_0",
     "params_b": 14,
     "load_gb": 15.9,
     "min_ram_gb": 24,
     "kv_kb_per_token": null,
     "ollama": "ollama run qwen3:14b-q8_0",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Coding, Quality",
     "mf_id": "qwen3-14b-q8"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "deepseek-r1-distill-qwen-14b",
   "name": "DeepSeek-R1 Distill Qwen 14B",
   "maker": "DeepSeek",
   "total_b": 14,
   "active_b": 14,
   "active_source": "dense",
   "arch": "dense",
   "use": "Reasoning, Quality",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "DeepSeek-R1 Distill Qwen 14B",
     "quant": "Q4_K_M",
     "params_b": 14,
     "load_gb": 11,
     "min_ram_gb": 16,
     "kv_kb_per_token": null,
     "ollama": "ollama run deepseek-r1:14b",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Reasoning, Quality",
     "mf_id": "deepseek-r1-distill-qwen-14b-q4"
    },
    {
     "label": "DeepSeek-R1 Distill Qwen 14B (Q8)",
     "quant": "Q8_0",
     "params_b": 14,
     "load_gb": 14.6,
     "min_ram_gb": 24,
     "kv_kb_per_token": null,
     "ollama": "ollama run deepseek-r1:14b-qwen-distill-q8_0",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Reasoning, Quality",
     "mf_id": "deepseek-r1-distill-qwen-14b-q8"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "ministral-3-14b",
   "name": "Ministral 3 14B",
   "maker": "Mistral",
   "total_b": 13.5,
   "active_b": null,
   "active_source": null,
   "arch": null,
   "use": "Chat, Reasoning, Vision",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Ministral 3 14B",
     "quant": "Q4_K_M",
     "params_b": 13.5,
     "load_gb": 9.1,
     "min_ram_gb": 16,
     "kv_kb_per_token": null,
     "ollama": "ollama run ministral-3:14b",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Chat, Reasoning, Vision",
     "mf_id": "ministral-3-14b-q4"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "mellum2.1-12b-a2.5b-thinking",
   "name": "Mellum2.1 12B-A2.5B Thinking",
   "maker": "Mellum",
   "total_b": 12.1,
   "active_b": 2.5,
   "active_source": "name",
   "arch": "moe",
   "use": "Agentic coding, Small machines",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Mellum2.1 12B-A2.5B Thinking",
     "quant": "Q4_K_M",
     "params_b": 12.1,
     "load_gb": 8.1,
     "min_ram_gb": 16,
     "kv_kb_per_token": 14,
     "ollama": null,
     "runtimes": [
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Agentic coding, Small machines",
     "mf_id": null
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "mistral-nemo-12b",
   "name": "Mistral Nemo 12B",
   "maker": "Mistral",
   "total_b": 12,
   "active_b": null,
   "active_source": null,
   "arch": null,
   "use": "Chat, Translation",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Mistral Nemo 12B",
     "quant": "Q4_K_M",
     "params_b": 12,
     "load_gb": 9.5,
     "min_ram_gb": 16,
     "kv_kb_per_token": null,
     "ollama": "ollama run mistral-nemo:12b",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Chat, Translation",
     "mf_id": "mistral-nemo-12b-q4"
    },
    {
     "label": "Mistral Nemo 12B (Q8)",
     "quant": "Q8_0",
     "params_b": 12,
     "load_gb": 12.1,
     "min_ram_gb": 24,
     "kv_kb_per_token": null,
     "ollama": "ollama run mistral-nemo:12b-instruct-2407-q8_0",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Chat, Translation",
     "mf_id": "mistral-nemo-12b-q8"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "gemma-3-12b-instruct",
   "name": "Gemma 3 12B Instruct",
   "maker": "Gemma",
   "total_b": 12,
   "active_b": 12,
   "active_source": "dense",
   "arch": "dense",
   "use": "Chat, Quality",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Gemma 3 12B Instruct",
     "quant": "Q4_K_M",
     "params_b": 12,
     "load_gb": 9.5,
     "min_ram_gb": 16,
     "kv_kb_per_token": null,
     "ollama": "ollama run gemma3:12b",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Chat, Quality",
     "mf_id": "gemma3-12b-q4"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "gemma-4-12b",
   "name": "Gemma 4 12B",
   "maker": "Gemma",
   "total_b": 12,
   "active_b": 12,
   "active_source": "dense",
   "arch": "dense",
   "use": "Chat, Coding, Multimodal",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Gemma 4 12B",
     "quant": "Q4_K_M",
     "params_b": 12,
     "load_gb": 8,
     "min_ram_gb": 12,
     "kv_kb_per_token": null,
     "ollama": "ollama run gemma4:12b",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Chat, Coding, Multimodal",
     "mf_id": "gemma4-12b-q4"
    },
    {
     "label": "Gemma 4 12B (Q8)",
     "quant": "Q8_0",
     "params_b": 12,
     "load_gb": 12.8,
     "min_ram_gb": 24,
     "kv_kb_per_token": null,
     "ollama": "ollama run gemma4:12b-it-q8_0",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Chat, Coding, Multimodal",
     "mf_id": "gemma4-12b-q8"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "glm-4-9b-chat",
   "name": "GLM-4-9B-Chat",
   "maker": "Zhipu",
   "total_b": 9.4,
   "active_b": null,
   "active_source": null,
   "arch": null,
   "use": "Chat, Translation, Tool calling",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "GLM-4-9B-Chat",
     "quant": "Q4_K_M",
     "params_b": 9.4,
     "load_gb": 6.3,
     "min_ram_gb": 12,
     "kv_kb_per_token": null,
     "ollama": "ollama run glm4:9b-chat-q4_K_M",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Chat, Translation, Tool calling",
     "mf_id": "glm-4-9b-chat-q4"
    },
    {
     "label": "GLM-4-9B-Chat (Q8)",
     "quant": "Q8_0",
     "params_b": 9.4,
     "load_gb": 10,
     "min_ram_gb": 16,
     "kv_kb_per_token": null,
     "ollama": "ollama run glm4:9b-chat-q8_0",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Chat, Translation, Tool calling",
     "mf_id": "glm-4-9b-chat-q8"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "gemma-2-9b-instruct",
   "name": "Gemma 2 9B Instruct",
   "maker": "Gemma",
   "total_b": 9,
   "active_b": 9,
   "active_source": "dense",
   "arch": "dense",
   "use": "Chat, Coding",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Gemma 2 9B Instruct",
     "quant": "Q4_K_M",
     "params_b": 9,
     "load_gb": 7,
     "min_ram_gb": 12,
     "kv_kb_per_token": null,
     "ollama": "ollama run gemma2:9b-instruct-q4_K_M",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Chat, Coding",
     "mf_id": "gemma2-9b-q4"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "qwen3.5-9b-instruct",
   "name": "Qwen3.5 9B Instruct",
   "maker": "Qwen",
   "total_b": 9,
   "active_b": 9,
   "active_source": "dense",
   "arch": "dense",
   "use": "Quality, Coding, Reasoning",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Qwen3.5 9B Instruct",
     "quant": "Q4_K_M",
     "params_b": 9,
     "load_gb": 7,
     "min_ram_gb": 12,
     "kv_kb_per_token": 32,
     "ollama": "ollama run qwen3.5:9b",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Quality, Coding, Reasoning",
     "mf_id": "qwen3.5-9b-q4"
    },
    {
     "label": "Qwen3.5 9B Instruct (Q8)",
     "quant": "Q8_0",
     "params_b": 9,
     "load_gb": 10.7,
     "min_ram_gb": 16,
     "kv_kb_per_token": 32,
     "ollama": "ollama run qwen3.5:9b-q8_0",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Quality, Coding, Reasoning",
     "mf_id": "qwen3.5-9b-q8"
    }
   ],
   "eci": 139.37,
   "shortlist": true
  },
  {
   "slug": "ornith-1.0-9b",
   "name": "Ornith 1.0 9B",
   "maker": "Ornith",
   "total_b": 9,
   "active_b": null,
   "active_source": null,
   "arch": null,
   "use": "Agentic coding on small machines",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Ornith 1.0 9B",
     "quant": "Q4_K_M",
     "params_b": 9,
     "load_gb": 5.6,
     "min_ram_gb": 8,
     "kv_kb_per_token": null,
     "ollama": "ollama run ornith:9b",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Agentic coding on small machines",
     "mf_id": "ornith-9b-q4"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "mimo-v2.6-distill-qwen-9b",
   "name": "MiMo-V2.6-Distill-Qwen-9B",
   "maker": "MiMo",
   "total_b": 9,
   "active_b": null,
   "active_source": null,
   "arch": null,
   "use": "Agentic coding, Small machines",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "MiMo-V2.6-Distill-Qwen-9B",
     "quant": "Q4_K_M",
     "params_b": 9,
     "load_gb": 5.8,
     "min_ram_gb": 16,
     "kv_kb_per_token": null,
     "ollama": null,
     "runtimes": [
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Agentic coding, Small machines",
     "mf_id": "mimo-v2.6-distill-qwen-9b"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "jebadiah-9b",
   "name": "Jebadiah 9B",
   "maker": "Jebadiah",
   "total_b": 9,
   "active_b": null,
   "active_source": null,
   "arch": null,
   "use": "Decisions",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Jebadiah 9B",
     "quant": "Q4_K_M",
     "params_b": 9,
     "load_gb": 5.8,
     "min_ram_gb": 12,
     "kv_kb_per_token": 32,
     "ollama": null,
     "runtimes": [
      "llama.cpp"
     ],
     "best_for": "Decisions",
     "mf_id": "jebadiah-9b-q4"
    },
    {
     "label": "Jebadiah 9B (Q5)",
     "quant": "Q5_K_M",
     "params_b": 9,
     "load_gb": 6.6,
     "min_ram_gb": 12,
     "kv_kb_per_token": 32,
     "ollama": null,
     "runtimes": [
      "llama.cpp"
     ],
     "best_for": "Decisions",
     "mf_id": "jebadiah-9b-q5"
    },
    {
     "label": "Jebadiah 9B (Q8)",
     "quant": "Q8_0",
     "params_b": 9,
     "load_gb": 9.8,
     "min_ram_gb": 16,
     "kv_kb_per_token": 32,
     "ollama": null,
     "runtimes": [
      "llama.cpp"
     ],
     "best_for": "Decisions",
     "mf_id": "jebadiah-9b-q8"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "granite-4.2-8b",
   "name": "Granite 4.2 8B",
   "maker": "Granite",
   "total_b": 8.8,
   "active_b": null,
   "active_source": null,
   "arch": null,
   "use": "Chat, Coding",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Granite 4.2 8B",
     "quant": "Q4_K_M",
     "params_b": 8.8,
     "load_gb": 5,
     "min_ram_gb": 8,
     "kv_kb_per_token": null,
     "ollama": "ollama run granite4.2:8b",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Chat, Coding",
     "mf_id": "granite4.2-8b-q4"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "granite-switch-4.1-8b",
   "name": "Granite Switch 4.1 8B",
   "maker": "Granite",
   "total_b": 8.8,
   "active_b": null,
   "active_source": null,
   "arch": null,
   "use": "Chat, Decisions, Safety checks",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Granite Switch 4.1 8B",
     "quant": "Q4_K_M",
     "params_b": 8.8,
     "load_gb": 6.4,
     "min_ram_gb": 12,
     "kv_kb_per_token": null,
     "ollama": null,
     "runtimes": [
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Chat, Decisions, Safety checks",
     "mf_id": "granite-switch-4.1-8b-q4"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "minicpm-v-4.5-8b",
   "name": "MiniCPM-V 4.5 8B",
   "maker": "MiniCPM",
   "total_b": 8.7,
   "active_b": null,
   "active_source": null,
   "arch": null,
   "use": "Vision, Multimodal",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "MiniCPM-V 4.5 8B",
     "quant": "Q4_K_M",
     "params_b": 8.7,
     "load_gb": 5.7,
     "min_ram_gb": 12,
     "kv_kb_per_token": null,
     "ollama": "ollama run minicpm-v4.5:8b",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Vision, Multimodal",
     "mf_id": "minicpm-v4.5-8b-q4"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "ministral-3-8b",
   "name": "Ministral 3 8B",
   "maker": "Mistral",
   "total_b": 8.4,
   "active_b": null,
   "active_source": null,
   "arch": null,
   "use": "Chat, Coding, Vision",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Ministral 3 8B",
     "quant": "Q4_K_M",
     "params_b": 8.4,
     "load_gb": 6,
     "min_ram_gb": 12,
     "kv_kb_per_token": null,
     "ollama": "ollama run ministral-3:8b",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Chat, Coding, Vision",
     "mf_id": "ministral-3-8b-q4"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "lfm2.5-8b-a1b",
   "name": "LFM2.5 8B-A1B",
   "maker": "LFM2",
   "total_b": 8.3,
   "active_b": 1.0,
   "active_source": "name",
   "arch": "moe",
   "use": "On-device agents, tool calling, multilingual chat",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "LFM2.5 8B-A1B",
     "quant": "Q4_K_M",
     "params_b": 8.3,
     "load_gb": 5.5,
     "min_ram_gb": 8,
     "kv_kb_per_token": null,
     "ollama": "ollama run lfm2.5:8b-a1b-q4_K_M",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "On-device agents, tool calling, multilingual chat",
     "mf_id": "lfm2.5-8b-a1b-q4"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "llama-3.1-8b-instruct",
   "name": "Llama 3.1 8B Instruct",
   "maker": "Llama",
   "total_b": 8,
   "active_b": 8,
   "active_source": "dense",
   "arch": "dense",
   "use": "Chat, Coding",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Llama 3.1 8B Instruct",
     "quant": "Q4_K_M",
     "params_b": 8,
     "load_gb": 6.5,
     "min_ram_gb": 12,
     "kv_kb_per_token": null,
     "ollama": "ollama run llama3.1:8b-instruct-q4_K_M",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Chat, Coding",
     "mf_id": "llama3.1-8b-q4"
    },
    {
     "label": "Llama 3.1 8B Instruct (Q8)",
     "quant": "Q8_0",
     "params_b": 8,
     "load_gb": 8,
     "min_ram_gb": 16,
     "kv_kb_per_token": null,
     "ollama": "ollama run llama3.1:8b-instruct-q8_0",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Chat, Coding",
     "mf_id": "llama3.1-8b-q8"
    },
    {
     "label": "Llama 3.1 8B Instruct (Q5)",
     "quant": "Q5_K_M",
     "params_b": 8,
     "load_gb": 8,
     "min_ram_gb": 12,
     "kv_kb_per_token": null,
     "ollama": "ollama run llama3.1:8b-instruct-q5_K_M",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Chat, Coding",
     "mf_id": "llama3.1-8b-q5"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "qwen3-8b",
   "name": "Qwen3 8B",
   "maker": "Qwen",
   "total_b": 8,
   "active_b": 8,
   "active_source": "dense",
   "arch": "dense",
   "use": "Chat, Coding",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Qwen3 8B",
     "quant": "Q4_K_M",
     "params_b": 8,
     "load_gb": 6.5,
     "min_ram_gb": 12,
     "kv_kb_per_token": null,
     "ollama": "ollama run qwen3:8b-q4_K_M",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Chat, Coding",
     "mf_id": "qwen3-8b-q4"
    },
    {
     "label": "Qwen3 8B (Q8)",
     "quant": "Q8_0",
     "params_b": 8,
     "load_gb": 8.1,
     "min_ram_gb": 16,
     "kv_kb_per_token": null,
     "ollama": "ollama run qwen3:8b-q8_0",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Chat, Coding",
     "mf_id": "qwen3-8b-q8"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "granite-4.1-8b-instruct",
   "name": "Granite 4.1 8B Instruct",
   "maker": "Granite",
   "total_b": 8,
   "active_b": null,
   "active_source": null,
   "arch": null,
   "use": "Enterprise assistant, tool calling, instruction following",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Granite 4.1 8B Instruct",
     "quant": "Q4_K_M",
     "params_b": 8,
     "load_gb": 5.5,
     "min_ram_gb": 8,
     "kv_kb_per_token": null,
     "ollama": "ollama run granite4.1:8b",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Enterprise assistant, tool calling, instruction following",
     "mf_id": "granite4.1-8b-q4"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "qwen2.5-7b-instruct",
   "name": "Qwen2.5 7B Instruct",
   "maker": "Qwen",
   "total_b": 7,
   "active_b": 7,
   "active_source": "dense",
   "arch": "dense",
   "use": "Chat, Coding",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Qwen2.5 7B Instruct",
     "quant": "Q4_K_M",
     "params_b": 7,
     "load_gb": 5.5,
     "min_ram_gb": 8,
     "kv_kb_per_token": null,
     "ollama": "ollama run qwen2.5:7b-instruct-q4_K_M",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Chat, Coding",
     "mf_id": "qwen2.5-7b-q4"
    }
   ],
   "eci": 118.46,
   "shortlist": true
  },
  {
   "slug": "mistral-7b-instruct",
   "name": "Mistral 7B Instruct",
   "maker": "Mistral",
   "total_b": 7,
   "active_b": 7,
   "active_source": "dense",
   "arch": "dense",
   "use": "Chat, Coding",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Mistral 7B Instruct",
     "quant": "Q4_K_M",
     "params_b": 7,
     "load_gb": 5.5,
     "min_ram_gb": 8,
     "kv_kb_per_token": null,
     "ollama": "ollama run mistral:7b-instruct-v0.3-q4_K_M",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Chat, Coding",
     "mf_id": "mistral-7b-q4"
    },
    {
     "label": "Mistral 7B Instruct (Q5)",
     "quant": "Q5_K_M",
     "params_b": 7,
     "load_gb": 4.8,
     "min_ram_gb": 8,
     "kv_kb_per_token": null,
     "ollama": "ollama run mistral:7b-instruct-v0.3-q5_K_M",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Chat, Coding",
     "mf_id": "mistral-7b-q5"
    },
    {
     "label": "Mistral 7B Instruct (Q8)",
     "quant": "Q8_0",
     "params_b": 7,
     "load_gb": 7.2,
     "min_ram_gb": 16,
     "kv_kb_per_token": null,
     "ollama": "ollama run mistral:7b-instruct-v0.3-q8_0",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Chat, Coding",
     "mf_id": "mistral-7b-q8"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "qwen2.5-coder-7b",
   "name": "Qwen2.5 Coder 7B",
   "maker": "Qwen",
   "total_b": 7,
   "active_b": 7,
   "active_source": "dense",
   "arch": "dense",
   "use": "Coding",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Qwen2.5 Coder 7B",
     "quant": "Q4_K_M",
     "params_b": 7,
     "load_gb": 5.5,
     "min_ram_gb": 8,
     "kv_kb_per_token": null,
     "ollama": "ollama run qwen2.5-coder:7b",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Coding",
     "mf_id": "qwen2.5-coder-7b-q4"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "deepseek-r1-distill-qwen-7b",
   "name": "DeepSeek-R1 Distill Qwen 7B",
   "maker": "DeepSeek",
   "total_b": 7,
   "active_b": 7,
   "active_source": "dense",
   "arch": "dense",
   "use": "Reasoning, Coding",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "DeepSeek-R1 Distill Qwen 7B",
     "quant": "Q4_K_M",
     "params_b": 7,
     "load_gb": 5.5,
     "min_ram_gb": 8,
     "kv_kb_per_token": null,
     "ollama": "ollama run deepseek-r1:7b",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Reasoning, Coding",
     "mf_id": "deepseek-r1-distill-qwen-7b-q4"
    },
    {
     "label": "DeepSeek-R1 Distill Qwen 7B (Q8)",
     "quant": "Q8_0",
     "params_b": 7,
     "load_gb": 7.5,
     "min_ram_gb": 16,
     "kv_kb_per_token": null,
     "ollama": "ollama run deepseek-r1:7b-qwen-distill-q8_0",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Reasoning, Coding",
     "mf_id": "deepseek-r1-distill-qwen-7b-q8"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "gemma-4-e4b",
   "name": "Gemma 4 E4B",
   "maker": "Gemma",
   "total_b": 4.5,
   "active_b": null,
   "active_source": null,
   "arch": null,
   "use": "On-device, Mobile, Chat",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Gemma 4 E4B",
     "quant": "Q4_K_M",
     "params_b": 4.5,
     "load_gb": 4,
     "min_ram_gb": 6,
     "kv_kb_per_token": null,
     "ollama": "ollama run gemma4:e4b",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "On-device, Mobile, Chat",
     "mf_id": "gemma4-e4b-q4"
    },
    {
     "label": "Gemma 4 E4B (Q8)",
     "quant": "Q8_0",
     "params_b": 4.5,
     "load_gb": 7.5,
     "min_ram_gb": 16,
     "kv_kb_per_token": null,
     "ollama": "ollama run gemma4:e4b-it-q8_0",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "On-device, Mobile, Chat",
     "mf_id": "gemma4-e4b-q8"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "gemma-3-4b-instruct",
   "name": "Gemma 3 4B Instruct",
   "maker": "Gemma",
   "total_b": 4,
   "active_b": 4,
   "active_source": "dense",
   "arch": "dense",
   "use": "Chat, Coding",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Gemma 3 4B Instruct",
     "quant": "Q4_K_M",
     "params_b": 4,
     "load_gb": 3.5,
     "min_ram_gb": 6,
     "kv_kb_per_token": null,
     "ollama": "ollama run gemma3:4b",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Chat, Coding",
     "mf_id": "gemma3-4b-q4"
    },
    {
     "label": "Gemma 3 4B Instruct (Q8)",
     "quant": "Q8_0",
     "params_b": 4,
     "load_gb": 3.9,
     "min_ram_gb": 8,
     "kv_kb_per_token": null,
     "ollama": "ollama run gemma3:4b-it-q8_0",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Chat, Coding",
     "mf_id": "gemma3-4b-q8"
    }
   ],
   "eci": 115.8,
   "shortlist": true
  },
  {
   "slug": "qwen3.5-4b-instruct",
   "name": "Qwen3.5 4B Instruct",
   "maker": "Qwen",
   "total_b": 4,
   "active_b": 4,
   "active_source": "dense",
   "arch": "dense",
   "use": "Coding, Agents, Multimodal",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Qwen3.5 4B Instruct",
     "quant": "Q4_K_M",
     "params_b": 4,
     "load_gb": 3.5,
     "min_ram_gb": 6,
     "kv_kb_per_token": 32,
     "ollama": "ollama run qwen3.5:4b",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Coding, Agents, Multimodal",
     "mf_id": "qwen3.5-4b-q4"
    },
    {
     "label": "Qwen3.5 4B Instruct (Q8)",
     "quant": "Q8_0",
     "params_b": 4,
     "load_gb": 4.3,
     "min_ram_gb": 8,
     "kv_kb_per_token": 32,
     "ollama": "ollama run qwen3.5:4b-q8_0",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Coding, Agents, Multimodal",
     "mf_id": "qwen3.5-4b-q8"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "nvidia-nemotron-3.5-bolt-e4b",
   "name": "NVIDIA Nemotron 3.5 Bolt E4B",
   "maker": "Nemotron",
   "total_b": 4,
   "active_b": null,
   "active_source": null,
   "arch": null,
   "use": "Voice assistants, Tool calling, Chat",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "NVIDIA Nemotron 3.5 Bolt E4B",
     "quant": "Q4_K_M",
     "params_b": 4,
     "load_gb": 7.9,
     "min_ram_gb": 12,
     "kv_kb_per_token": null,
     "ollama": "ollama run nemotron-3.5-bolt:e4b",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Voice assistants, Tool calling, Chat",
     "mf_id": "nemotron-3.5-bolt-e4b-q4"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "phi-3-mini-3.8b",
   "name": "Phi-3 Mini 3.8B",
   "maker": "Phi",
   "total_b": 3.8,
   "active_b": 3.8,
   "active_source": "dense",
   "arch": "dense",
   "use": "Coding, Chat",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Phi-3 Mini 3.8B",
     "quant": "Q4_K_M",
     "params_b": 3.8,
     "load_gb": 3.2,
     "min_ram_gb": 6,
     "kv_kb_per_token": null,
     "ollama": "ollama run phi3:3.8b-mini-4k-instruct-q4_K_M",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Coding, Chat",
     "mf_id": null
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "phi-4-mini-3.8b",
   "name": "Phi-4 Mini 3.8B",
   "maker": "Phi",
   "total_b": 3.8,
   "active_b": 3.8,
   "active_source": "dense",
   "arch": "dense",
   "use": "Coding, Chat",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Phi-4 Mini 3.8B",
     "quant": "Q4_K_M",
     "params_b": 3.8,
     "load_gb": 3.2,
     "min_ram_gb": 6,
     "kv_kb_per_token": null,
     "ollama": "ollama run phi4-mini:3.8b",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Coding, Chat",
     "mf_id": "phi4-mini-3.8b-q4"
    },
    {
     "label": "Phi-4 Mini 3.8B (Q8)",
     "quant": "Q8_0",
     "params_b": 3.8,
     "load_gb": 3.8,
     "min_ram_gb": 8,
     "kv_kb_per_token": null,
     "ollama": "ollama run phi4-mini:3.8b-q8_0",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Coding, Chat",
     "mf_id": "phi4-mini-3.8b-q8"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "granite-4.2-3b",
   "name": "Granite 4.2 3B",
   "maker": "Granite",
   "total_b": 3.7,
   "active_b": null,
   "active_source": null,
   "arch": null,
   "use": "Chat, Coding",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Granite 4.2 3B",
     "quant": "Q4_K_M",
     "params_b": 3.7,
     "load_gb": 2.1,
     "min_ram_gb": 4,
     "kv_kb_per_token": null,
     "ollama": "ollama run granite4.2:3b",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Chat, Coding",
     "mf_id": "granite4.2-3b-q4"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "ministral-3-3b",
   "name": "Ministral 3 3B",
   "maker": "Mistral",
   "total_b": 3.4,
   "active_b": null,
   "active_source": null,
   "arch": null,
   "use": "Chat, Vision, Edge",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Ministral 3 3B",
     "quant": "Q4_K_M",
     "params_b": 3.4,
     "load_gb": 3,
     "min_ram_gb": 6,
     "kv_kb_per_token": null,
     "ollama": "ollama run ministral-3:3b",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Chat, Vision, Edge",
     "mf_id": "ministral-3-3b-q4"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "qwen2.5-3b-instruct",
   "name": "Qwen2.5 3B Instruct",
   "maker": "Qwen",
   "total_b": 3,
   "active_b": 3,
   "active_source": "dense",
   "arch": "dense",
   "use": "Chat, Coding",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Qwen2.5 3B Instruct",
     "quant": "Q4_K_M",
     "params_b": 3,
     "load_gb": 2.5,
     "min_ram_gb": 4,
     "kv_kb_per_token": null,
     "ollama": "ollama run qwen2.5:3b-instruct-q4_K_M",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Chat, Coding",
     "mf_id": "qwen2.5-3b-q4"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "llama-3.2-3b-instruct",
   "name": "Llama 3.2 3B Instruct",
   "maker": "Llama",
   "total_b": 3,
   "active_b": 3,
   "active_source": "dense",
   "arch": "dense",
   "use": "Chat",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Llama 3.2 3B Instruct",
     "quant": "Q4_K_M",
     "params_b": 3,
     "load_gb": 2.5,
     "min_ram_gb": 4,
     "kv_kb_per_token": null,
     "ollama": "ollama run llama3.2:3b-instruct-q4_K_M",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Chat",
     "mf_id": "llama3.2-3b-q4"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "granite-4.1-3b-instruct",
   "name": "Granite 4.1 3B Instruct",
   "maker": "Granite",
   "total_b": 3,
   "active_b": null,
   "active_source": null,
   "arch": null,
   "use": "Lightweight chat, classification, edge tasks",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Granite 4.1 3B Instruct",
     "quant": "Q4_K_M",
     "params_b": 3,
     "load_gb": 2,
     "min_ram_gb": 4,
     "kv_kb_per_token": null,
     "ollama": "ollama run granite4.1:3b",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Lightweight chat, classification, edge tasks",
     "mf_id": "granite4.1-3b-q4"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "gemma-4-e2b",
   "name": "Gemma 4 E2B",
   "maker": "Gemma",
   "total_b": 2.3,
   "active_b": null,
   "active_source": null,
   "arch": null,
   "use": "IoT, Mobile, Edge",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Gemma 4 E2B",
     "quant": "Q4_K_M",
     "params_b": 2.3,
     "load_gb": 2.3,
     "min_ram_gb": 4,
     "kv_kb_per_token": null,
     "ollama": "ollama run gemma4:e2b",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "IoT, Mobile, Edge",
     "mf_id": "gemma4-e2b-q4"
    },
    {
     "label": "Gemma 4 E2B (Q8)",
     "quant": "Q8_0",
     "params_b": 2.3,
     "load_gb": 4.6,
     "min_ram_gb": 8,
     "kv_kb_per_token": null,
     "ollama": "ollama run gemma4:e2b-it-q8_0",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "IoT, Mobile, Edge",
     "mf_id": "gemma4-e2b-q8"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "gemma-2-2b-instruct",
   "name": "Gemma 2 2B Instruct",
   "maker": "Gemma",
   "total_b": 2,
   "active_b": 2,
   "active_source": "dense",
   "arch": "dense",
   "use": "Chat",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Gemma 2 2B Instruct",
     "quant": "Q4_K_M",
     "params_b": 2,
     "load_gb": 1.8,
     "min_ram_gb": 4,
     "kv_kb_per_token": null,
     "ollama": "ollama run gemma2:2b-instruct-q4_K_M",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Chat",
     "mf_id": "gemma2-2b-q4"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "qwen3.5-2b-instruct",
   "name": "Qwen3.5 2B Instruct",
   "maker": "Qwen",
   "total_b": 2,
   "active_b": 2,
   "active_source": "dense",
   "arch": "dense",
   "use": "Chat, Edge tasks",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Qwen3.5 2B Instruct",
     "quant": "Q4_K_M",
     "params_b": 2,
     "load_gb": 1.8,
     "min_ram_gb": 4,
     "kv_kb_per_token": 12,
     "ollama": "ollama run qwen3.5:2b",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Chat, Edge tasks",
     "mf_id": "qwen3.5-2b-q4"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "lfm3-1.7b",
   "name": "LFM3 1.7B",
   "maker": "LFM3",
   "total_b": 1.73,
   "active_b": null,
   "active_source": null,
   "arch": null,
   "use": "Local agents, Tool calling, Structured output",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "LFM3 1.7B",
     "quant": "Q4_K_M",
     "params_b": 1.73,
     "load_gb": 1.1,
     "min_ram_gb": 4,
     "kv_kb_per_token": 12,
     "ollama": "ollama run lfm3:1.7b",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Local agents, Tool calling, Structured output",
     "mf_id": "lfm3-1.7b-q4"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "qwen2.5-1.5b-instruct",
   "name": "Qwen2.5 1.5B Instruct",
   "maker": "Qwen",
   "total_b": 1.5,
   "active_b": 1.5,
   "active_source": "dense",
   "arch": "dense",
   "use": "Chat, Translation",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Qwen2.5 1.5B Instruct",
     "quant": "Q4_K_M",
     "params_b": 1.5,
     "load_gb": 1.5,
     "min_ram_gb": 4,
     "kv_kb_per_token": null,
     "ollama": "ollama run qwen2.5:1.5b-instruct-q4_K_M",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Chat, Translation",
     "mf_id": "qwen2.5-1.5b-q4"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "llama-3.2-1b-instruct",
   "name": "Llama 3.2 1B Instruct",
   "maker": "Llama",
   "total_b": 1,
   "active_b": 1,
   "active_source": "dense",
   "arch": "dense",
   "use": "Chat",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Llama 3.2 1B Instruct",
     "quant": "Q4_K_M",
     "params_b": 1,
     "load_gb": 1,
     "min_ram_gb": 2,
     "kv_kb_per_token": null,
     "ollama": "ollama run llama3.2:1b-instruct-q4_K_M",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Chat",
     "mf_id": "llama3.2-1b-q4"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "gemma-3-1b-instruct",
   "name": "Gemma 3 1B Instruct",
   "maker": "Gemma",
   "total_b": 1,
   "active_b": 1,
   "active_source": "dense",
   "arch": "dense",
   "use": "Chat, Mobile",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Gemma 3 1B Instruct",
     "quant": "Q4_K_M",
     "params_b": 1,
     "load_gb": 1,
     "min_ram_gb": 2,
     "kv_kb_per_token": null,
     "ollama": "ollama run gemma3:1b",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Chat, Mobile",
     "mf_id": "gemma-3-1b-q4"
    },
    {
     "label": "Gemma 3 1B Instruct (Q8)",
     "quant": "Q8_0",
     "params_b": 1,
     "load_gb": 1,
     "min_ram_gb": 4,
     "kv_kb_per_token": null,
     "ollama": "ollama run gemma3:1b-it-q8_0",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Chat, Mobile",
     "mf_id": "gemma-3-1b-q8"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "qwen3.5-0.8b-instruct",
   "name": "Qwen3.5 0.8B Instruct",
   "maker": "Qwen",
   "total_b": 0.8,
   "active_b": 0.8,
   "active_source": "dense",
   "arch": "dense",
   "use": "Chat, Mobile",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Qwen3.5 0.8B Instruct",
     "quant": "Q4_K_M",
     "params_b": 0.8,
     "load_gb": 0.8,
     "min_ram_gb": 2,
     "kv_kb_per_token": 12,
     "ollama": "ollama run qwen3.5:0.8b",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Chat, Mobile",
     "mf_id": "qwen3.5-0.8b-q4"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "qwen2.5-0.5b-instruct",
   "name": "Qwen2.5 0.5B Instruct",
   "maker": "Qwen",
   "total_b": 0.5,
   "active_b": 0.5,
   "active_source": "dense",
   "arch": "dense",
   "use": "Chat, Mobile",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "Qwen2.5 0.5B Instruct",
     "quant": "Q4_K_M",
     "params_b": 0.5,
     "load_gb": 0.8,
     "min_ram_gb": 2,
     "kv_kb_per_token": null,
     "ollama": "ollama run qwen2.5:0.5b-instruct-q4_K_M",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Chat, Mobile",
     "mf_id": "qwen2.5-0.5b-q4"
    }
   ],
   "eci": null,
   "shortlist": false
  },
  {
   "slug": "smollm2-360m",
   "name": "SmolLM2 360M",
   "maker": "SmolLM",
   "total_b": 0.36,
   "active_b": null,
   "active_source": null,
   "arch": null,
   "use": "Chat, Embedded",
   "source": null,
   "checked": null,
   "gpuwerk": null,
   "spark": null,
   "hf_check": null,
   "variants": [
    {
     "label": "SmolLM2 360M",
     "quant": "Q4_K_M",
     "params_b": 0.36,
     "load_gb": 0.5,
     "min_ram_gb": 1,
     "kv_kb_per_token": null,
     "ollama": "ollama run smollm2:360m",
     "runtimes": [
      "ollama",
      "llama.cpp",
      "lm-studio"
     ],
     "best_for": "Chat, Embedded",
     "mf_id": "smollm2-360m-q4"
    }
   ],
   "eci": null,
   "shortlist": false
  }
 ]
}