{
  "version": 2,
  "models": [
    {
      "id": "qwen3.5-2b-q4",
      "name": "Qwen 3.5 2B",
      "family": "qwen",
      "gguf_file": "Qwen3.5-2B-Q4_K_M.gguf",
      "gguf_url": "https://huggingface.co/unsloth/Qwen3.5-2B-GGUF/resolve/main/Qwen3.5-2B-Q4_K_M.gguf",
      "gguf_sha256": "aaf42c8b7c3cab2bf3d69c355048d4a0ee9973d48f16c731c0520ee914699223",
      "size_mb": 1500,
      "vram_required_gb": 3,
      "context_length": 8192,
      "quantization": "Q4_K_M",
      "specialty": "Fast",
      "description": "Lightweight model for quick responses. Best for simple tasks and low-resource systems.",
      "tokens_per_sec_estimate": 120,
      "llm_model_name": "qwen3.5-2b",
      "llama_server_image": null
    },
    {
      "id": "phi4-mini-q4",
      "name": "Phi-4 Mini",
      "family": "phi",
      "gguf_file": "Phi-4-mini-instruct-Q4_K_M.gguf",
      "gguf_url": "https://huggingface.co/unsloth/Phi-4-mini-instruct-GGUF/resolve/main/Phi-4-mini-instruct-Q4_K_M.gguf",
      "gguf_sha256": "88c00229914083cd112853aab84ed51b87bdf6b9ce42f532d8c85c7c63b1730a",
      "size_mb": 2490,
      "vram_required_gb": 4,
      "context_length": 128000,
      "quantization": "Q4_K_M",
      "specialty": "Balanced",
      "description": "Microsoft's compact model with 128K context. Punches above its weight on reasoning and code.",
      "tokens_per_sec_estimate": 130,
      "llm_model_name": "phi-4-mini",
      "llama_server_image": null
    },
    {
      "id": "qwen3.5-4b-q4",
      "name": "Qwen 3.5 4B",
      "family": "qwen",
      "gguf_file": "Qwen3.5-4B-Q4_K_M.gguf",
      "gguf_url": "https://huggingface.co/unsloth/Qwen3.5-4B-GGUF/resolve/main/Qwen3.5-4B-Q4_K_M.gguf",
      "gguf_sha256": "00fe7986ff5f6b463e62455821146049db6f9313603938a70800d1fb69ef11a4",
      "size_mb": 2870,
      "vram_required_gb": 5,
      "context_length": 16384,
      "quantization": "Q4_K_M",
      "specialty": "Balanced",
      "description": "Good balance of speed and capability for Intel Arc and budget GPUs.",
      "tokens_per_sec_estimate": 100,
      "llm_model_name": "qwen3.5-4b",
      "llama_server_image": null
    },
    {
      "id": "gemma4-e2b-q4",
      "name": "Gemma 4 E2B",
      "family": "gemma4",
      "gguf_file": "gemma-4-E2B-it-Q4_K_M.gguf",
      "gguf_url": "https://huggingface.co/unsloth/gemma-4-E2B-it-GGUF/resolve/main/gemma-4-E2B-it-Q4_K_M.gguf",
      "gguf_sha256": "ac0069ebccd39925d836f24a88c0f0c858d20578c29b21ab7cedce66ee576845",
      "size_mb": 2810,
      "vram_required_gb": 5,
      "context_length": 16384,
      "quantization": "Q4_K_M",
      "specialty": "Fast",
      "description": "Google's efficient small model. Good for quick tasks on budget hardware.",
      "tokens_per_sec_estimate": 110,
      "llm_model_name": "gemma-4-e2b-it",
      "llama_server_image": "ghcr.io/ggml-org/llama.cpp:server-cuda-b9014"
    },
    {
      "id": "deepseek-r1-7b-q4",
      "name": "DeepSeek R1 7B",
      "family": "deepseek",
      "gguf_file": "DeepSeek-R1-Distill-Qwen-7B-Q4_K_M.gguf",
      "gguf_url": "https://huggingface.co/unsloth/DeepSeek-R1-Distill-Qwen-7B-GGUF/resolve/main/DeepSeek-R1-Distill-Qwen-7B-Q4_K_M.gguf",
      "gguf_sha256": "78272d8d32084548bd450394a560eb2d70de8232ab96a725769b1f9171235c1c",
      "size_mb": 4680,
      "vram_required_gb": 7,
      "context_length": 32768,
      "quantization": "Q4_K_M",
      "specialty": "Reasoning",
      "description": "DeepSeek R1 distilled into Qwen 7B. Chain-of-thought reasoning on a budget GPU.",
      "tokens_per_sec_estimate": 80,
      "llm_model_name": "deepseek-r1-distill-qwen-7b",
      "llama_server_image": null
    },
    {
      "id": "gemma4-e4b-q4",
      "name": "Gemma 4 E4B",
      "family": "gemma4",
      "gguf_file": "gemma-4-E4B-it-Q4_K_M.gguf",
      "gguf_url": "https://huggingface.co/unsloth/gemma-4-E4B-it-GGUF/resolve/main/gemma-4-E4B-it-Q4_K_M.gguf",
      "gguf_sha256": "dff0ffba4c90b4082d70214d53ce9504a28d4d8d998276dcb3b8881a656c742a",
      "size_mb": 5340,
      "vram_required_gb": 8,
      "context_length": 32768,
      "quantization": "Q4_K_M",
      "specialty": "General",
      "description": "Google's mid-range model with strong multilingual and reasoning capabilities.",
      "tokens_per_sec_estimate": 85,
      "llm_model_name": "gemma-4-e4b-it",
      "llama_server_image": "ghcr.io/ggml-org/llama.cpp:server-cuda-b9014"
    },
    {
      "id": "qwen3.5-9b-q4",
      "name": "Qwen 3.5 9B",
      "family": "qwen",
      "gguf_file": "Qwen3.5-9B-Q4_K_M.gguf",
      "gguf_url": "https://huggingface.co/unsloth/Qwen3.5-9B-GGUF/resolve/main/Qwen3.5-9B-Q4_K_M.gguf",
      "gguf_sha256": "03b74727a860a56338e042c4420bb3f04b2fec5734175f4cb9fa853daf52b7e8",
      "size_mb": 5760,
      "vram_required_gb": 8,
      "context_length": 32768,
      "quantization": "Q4_K_M",
      "specialty": "General",
      "description": "Excellent general-purpose model. Default for most systems. Great at coding, reasoning, and multilingual tasks.",
      "tokens_per_sec_estimate": 90,
      "llm_model_name": "qwen3.5-9b",
      "llama_server_image": null
    },
    {
      "id": "phi4-q4",
      "name": "Phi-4 14B",
      "family": "phi",
      "gguf_file": "phi-4-Q4_K_M.gguf",
      "gguf_url": "https://huggingface.co/bartowski/phi-4-GGUF/resolve/main/phi-4-Q4_K_M.gguf",
      "gguf_sha256": "009aba717c09d4a35890c7d35eb59d54e1dba884c7c526e7197d9c13ab5911d9",
      "size_mb": 9050,
      "vram_required_gb": 11,
      "context_length": 16384,
      "quantization": "Q4_K_M",
      "specialty": "General",
      "description": "Microsoft's MIT-licensed 14B model. Strong at STEM, coding, and structured reasoning.",
      "tokens_per_sec_estimate": 35,
      "llm_model_name": "phi-4",
      "llama_server_image": null
    },
    {
      "id": "deepseek-r1-14b-q4",
      "name": "DeepSeek R1 14B",
      "family": "deepseek",
      "gguf_file": "DeepSeek-R1-Distill-Qwen-14B-Q4_K_M.gguf",
      "gguf_url": "https://huggingface.co/unsloth/DeepSeek-R1-Distill-Qwen-14B-GGUF/resolve/main/DeepSeek-R1-Distill-Qwen-14B-Q4_K_M.gguf",
      "gguf_sha256": "67a7933cf2ad596a393c8e13b30bc4da2d50b283e250b78554aed18817eca31c",
      "size_mb": 8990,
      "vram_required_gb": 12,
      "context_length": 32768,
      "quantization": "Q4_K_M",
      "specialty": "Reasoning",
      "description": "DeepSeek R1 distilled into Qwen 14B. Strong chain-of-thought reasoning for mid-range GPUs.",
      "tokens_per_sec_estimate": 50,
      "llm_model_name": "deepseek-r1-distill-qwen-14b",
      "llama_server_image": null
    },
    {
      "id": "qwen3.5-27b-q4",
      "name": "Qwen 3.5 27B",
      "family": "qwen",
      "gguf_file": "Qwen3.5-27B-Q4_K_M.gguf",
      "gguf_url": "https://huggingface.co/unsloth/Qwen3.5-27B-GGUF/resolve/main/Qwen3.5-27B-Q4_K_M.gguf",
      "gguf_sha256": "84b5f7f112156d63836a01a69dc3f11a6ba63b10a23b8ca7a7efaf52d5a2d806",
      "size_mb": 16700,
      "vram_required_gb": 20,
      "context_length": 32768,
      "quantization": "Q4_K_M",
      "specialty": "Quality",
      "description": "Dense 27B powerhouse. Top-tier quality for 24GB GPUs. Excels at complex reasoning and coding.",
      "tokens_per_sec_estimate": 25,
      "llm_model_name": "qwen3.5-27b",
      "llama_server_image": null
    },
    {
      "id": "gemma4-26b-a4b-q4",
      "name": "Gemma 4 26B-A4B",
      "family": "gemma4",
      "gguf_file": "gemma-4-26B-A4B-it-Q4_K_M.gguf",
      "gguf_url": "https://huggingface.co/ggml-org/gemma-4-26B-A4B-it-GGUF/resolve/main/gemma-4-26B-A4B-it-Q4_K_M.gguf",
      "gguf_sha256": "23c6997912cb7fa36147fe05877de73ddbb2a80ff69b18ff171b354dccf2b5b5",
      "size_mb": 18000,
      "vram_required_gb": 22,
      "context_length": 16384,
      "quantization": "Q4_K_M",
      "specialty": "Reasoning",
      "description": "Google's large MoE model. Strong at reasoning and analysis tasks.",
      "tokens_per_sec_estimate": 50,
      "llm_model_name": "gemma-4-26b-a4b-it",
      "llama_server_image": "ghcr.io/ggml-org/llama.cpp:server-cuda-b9014"
    },
    {
      "id": "qwen3-30b-a3b-q4",
      "name": "Qwen 3 30B-A3B",
      "family": "qwen",
      "gguf_file": "Qwen3-30B-A3B-Q4_K_M.gguf",
      "gguf_url": "https://huggingface.co/unsloth/Qwen3-30B-A3B-GGUF/resolve/main/Qwen3-30B-A3B-Q4_K_M.gguf",
      "gguf_sha256": "9f1a24700a339b09c06009b729b5c809e0b64c213b8af5b711b3dbdfd0c5ba48",
      "size_mb": 18600,
      "vram_required_gb": 22,
      "context_length": 131072,
      "quantization": "Q4_K_M",
      "specialty": "Quality",
      "description": "High-quality MoE model with 131K context. Excellent for complex reasoning and long documents.",
      "tokens_per_sec_estimate": 55,
      "llm_model_name": "qwen3-30b-a3b",
      "llama_server_image": null
    },
    {
      "id": "gemma4-31b-q4",
      "name": "Gemma 4 31B",
      "family": "gemma4",
      "gguf_file": "gemma-4-31B-it-Q4_K_M.gguf",
      "gguf_url": "https://huggingface.co/ggml-org/gemma-4-31B-it-GGUF/resolve/main/gemma-4-31B-it-Q4_K_M.gguf",
      "gguf_sha256": "a20deaf2f8fc27c501f32fadfd538f8f31a76f10f47d5d3eb895f3d1112d752c",
      "size_mb": 19800,
      "vram_required_gb": 24,
      "context_length": 131072,
      "quantization": "Q4_K_M",
      "specialty": "Quality",
      "description": "Google's flagship dense model with 131K context. Top-tier quality for enterprise GPUs.",
      "tokens_per_sec_estimate": 40,
      "llm_model_name": "gemma-4-31b-it",
      "llama_server_image": "ghcr.io/ggml-org/llama.cpp:server-cuda-b9014"
    },
    {
      "id": "deepseek-r1-32b-q4",
      "name": "DeepSeek R1 32B",
      "family": "deepseek",
      "gguf_file": "DeepSeek-R1-Distill-Qwen-32B-Q4_K_M.gguf",
      "gguf_url": "https://huggingface.co/unsloth/DeepSeek-R1-Distill-Qwen-32B-GGUF/resolve/main/DeepSeek-R1-Distill-Qwen-32B-Q4_K_M.gguf",
      "gguf_sha256": "ca171ca03554ee20cf67ad6b540610ae7eabb95af00c0abd36bb73542e140fb5",
      "size_mb": 19900,
      "vram_required_gb": 24,
      "context_length": 32768,
      "quantization": "Q4_K_M",
      "specialty": "Reasoning",
      "description": "DeepSeek R1 distilled into Qwen 32B. The best open-source reasoning model at this size.",
      "tokens_per_sec_estimate": 25,
      "llm_model_name": "deepseek-r1-distill-qwen-32b",
      "llama_server_image": null
    },
    {
      "id": "qwen3.5-35b-a3b-q4",
      "name": "Qwen 3.5 35B-A3B",
      "family": "qwen",
      "gguf_file": "Qwen3.5-35B-A3B-Q4_K_M.gguf",
      "gguf_url": "https://huggingface.co/unsloth/Qwen3.5-35B-A3B-GGUF/resolve/main/Qwen3.5-35B-A3B-Q4_K_M.gguf",
      "gguf_sha256": "3b46d1066bc91cc2d613e3bc22ce691dd77e6f0d33c9060690d24ce6de494375",
      "size_mb": 22000,
      "vram_required_gb": 24,
      "context_length": 131072,
      "quantization": "Q4_K_M",
      "specialty": "Quality",
      "description": "Successor to Qwen3 30B-A3B. MoE with 131K context — top scores on coding, math, and reasoning benchmarks.",
      "tokens_per_sec_estimate": 50,
      "llm_model_name": "qwen3.5-35b-a3b",
      "llama_server_image": null
    },
    {
      "id": "qwen3.6-35b-a3b-ud-q4",
      "name": "Qwen 3.6 35B-A3B",
      "family": "qwen",
      "gguf_file": "Qwen3.6-35B-A3B-UD-Q4_K_M.gguf",
      "gguf_url": "https://huggingface.co/unsloth/Qwen3.6-35B-A3B-GGUF/resolve/main/Qwen3.6-35B-A3B-UD-Q4_K_M.gguf",
      "gguf_sha256": "ac0e2c1189e055faa36eff361580e79c5bd6f8e76bffb4ce547f167d53e31a61",
      "size_mb": 21110,
      "vram_required_gb": 24,
      "context_length": 131072,
      "quantization": "UD-Q4_K_M",
      "specialty": "Quality",
      "description": "Current A3B MoE successor for Spark-class unified-memory hosts. Verified on DGX Spark with llama.cpp b9014.",
      "tokens_per_sec_estimate": 59,
      "llm_model_name": "qwen3.6-35b-a3b",
      "llama_server_image": null
    },
    {
      "id": "deepseek-r1-70b-q4",
      "name": "DeepSeek R1 70B",
      "family": "deepseek",
      "gguf_file": "DeepSeek-R1-Distill-Llama-70B-Q4_K_M.gguf",
      "gguf_url": "https://huggingface.co/unsloth/DeepSeek-R1-Distill-Llama-70B-GGUF/resolve/main/DeepSeek-R1-Distill-Llama-70B-Q4_K_M.gguf",
      "gguf_sha256": "952ff479c48ac3ece81fb6d9a5a03bfeac215e6caac780fbe91ff8cb0e05bcf3",
      "size_mb": 42500,
      "vram_required_gb": 48,
      "context_length": 32768,
      "quantization": "Q4_K_M",
      "specialty": "Reasoning",
      "description": "The largest DeepSeek R1 distill. State-of-the-art reasoning for dual-GPU or 48GB+ setups.",
      "tokens_per_sec_estimate": 15,
      "llm_model_name": "deepseek-r1-distill-llama-70b",
      "llama_server_image": null
    },
    {
      "id": "qwen3-coder-next-q4",
      "name": "Qwen 3 Coder Next",
      "family": "qwen",
      "gguf_file": "qwen3-coder-next-Q4_K_M.gguf",
      "gguf_url": "https://huggingface.co/unsloth/Qwen3-Coder-Next-GGUF/resolve/main/Qwen3-Coder-Next-Q4_K_M.gguf",
      "gguf_sha256": "9e6032d2f3b50a60f17ce8bf5a1d85c71af9b53b89c7978020ae7c660f29b090",
      "size_mb": 48500,
      "vram_required_gb": 52,
      "context_length": 131072,
      "quantization": "Q4_K_M",
      "specialty": "Code",
      "description": "Flagship coding model for enterprise GPUs and Strix Halo. 131K context for large codebases.",
      "tokens_per_sec_estimate": 30,
      "llm_model_name": "qwen3-coder-next",
      "llama_server_image": null
    },
    {
      "id": "llama4-scout-q4",
      "name": "Llama 4 Scout",
      "family": "llama4",
      "gguf_file": "Llama-4-Scout-17B-16E-Instruct-Q4_K_M-00001-of-00002.gguf",
      "gguf_url": "",
      "gguf_sha256": "",
      "gguf_parts": [
        {
          "file": "Llama-4-Scout-17B-16E-Instruct-Q4_K_M-00001-of-00002.gguf",
          "url": "https://huggingface.co/unsloth/Llama-4-Scout-17B-16E-Instruct-GGUF/resolve/main/Q4_K_M/Llama-4-Scout-17B-16E-Instruct-Q4_K_M-00001-of-00002.gguf",
          "sha256": "fbe956902467171ed7c0c326e5d868771a84d46468d407abecd0f289297313f9"
        },
        {
          "file": "Llama-4-Scout-17B-16E-Instruct-Q4_K_M-00002-of-00002.gguf",
          "url": "https://huggingface.co/unsloth/Llama-4-Scout-17B-16E-Instruct-GGUF/resolve/main/Q4_K_M/Llama-4-Scout-17B-16E-Instruct-Q4_K_M-00002-of-00002.gguf",
          "sha256": "e7330ae14527f08e8a44a6a573b45084052b8822c4b8a5179c5cad9cd6f6f795"
        }
      ],
      "size_mb": 65300,
      "vram_required_gb": 70,
      "context_length": 131072,
      "quantization": "Q4_K_M",
      "specialty": "General",
      "description": "Meta's flagship MoE with 16 experts. 131K context, strong multimodal capabilities. Needs 80GB+ VRAM.",
      "tokens_per_sec_estimate": 20,
      "llm_model_name": "llama-4-scout",
      "llama_server_image": null
    },
    {
      "id": "qwen3.5-122b-a10b-q4",
      "name": "Qwen 3.5 122B-A10B",
      "family": "qwen",
      "gguf_file": "Qwen3.5-122B-A10B-Q4_K_M-00001-of-00003.gguf",
      "gguf_url": "",
      "gguf_sha256": "",
      "gguf_parts": [
        {
          "file": "Qwen3.5-122B-A10B-Q4_K_M-00001-of-00003.gguf",
          "url": "https://huggingface.co/unsloth/Qwen3.5-122B-A10B-GGUF/resolve/main/Q4_K_M/Qwen3.5-122B-A10B-Q4_K_M-00001-of-00003.gguf",
          "sha256": "467c9bd92ea518539cf75bf5a5fbfbd35e9a0b40d766ccaa67bf120e12041df3"
        },
        {
          "file": "Qwen3.5-122B-A10B-Q4_K_M-00002-of-00003.gguf",
          "url": "https://huggingface.co/unsloth/Qwen3.5-122B-A10B-GGUF/resolve/main/Q4_K_M/Qwen3.5-122B-A10B-Q4_K_M-00002-of-00003.gguf",
          "sha256": "90db14846413aebdac365b57206441437cac5f7e5037d94b325f0167f902e6e7"
        },
        {
          "file": "Qwen3.5-122B-A10B-Q4_K_M-00003-of-00003.gguf",
          "url": "https://huggingface.co/unsloth/Qwen3.5-122B-A10B-GGUF/resolve/main/Q4_K_M/Qwen3.5-122B-A10B-Q4_K_M-00003-of-00003.gguf",
          "sha256": "e3c24b8ebec070bb4f69ea0aca25a16531da7440cd515529953e046882901f97"
        }
      ],
      "size_mb": 76500,
      "vram_required_gb": 80,
      "context_length": 131072,
      "quantization": "Q4_K_M",
      "specialty": "Quality",
      "description": "The largest Qwen 3.5 MoE. 122B params, 10B active. Top-tier quality for multi-GPU or 96GB+ unified memory systems.",
      "tokens_per_sec_estimate": 15,
      "llm_model_name": "qwen3.5-122b-a10b",
      "llama_server_image": null
    }
  ]
}