Create models.json

models.json

@@ -0,0 +1,1593 @@
+{
+  "name": "ModelFit Local LLM Hardware Compatibility Dataset",
+  "description": "Which local AI models (Ollama) fit which hardware, by parameter size, quantization, and estimated memory load. tok/s and load figures are estimates, not measured benchmarks.",
+  "source": "https://modelfit.io/data/",
+  "license": "CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/)",
+  "attribution": "ModelFit: https://modelfit.io/",
+  "methodology": {
+    "ramBudgetRatioBase": 0.7,
+    "ramBudgetRatioMax": 0.85,
+    "note": "A model fits a device when its estimatedLoadGb is within the memory budget: ~70% of unified memory up to 32GB, scaling linearly to ~85% at 128GB and above (high-RAM Macs can wire more memory to the GPU). Q4_K_M needs roughly 0.6 GB per billion parameters."
+  },
+  "updated": "2026-07-02",
+  "counts": {
+    "models": 106,
+    "local": 75,
+    "cloud": 31,
+    "families": 20
+  },
+  "models": [
+    {
+      "model": "Qwen2.5 1.5B Instruct",
+      "family": "Qwen",
+      "params": 1.5,
+      "quantization": "Q4_K_M",
+      "minRamGb": 4,
+      "estimatedLoadGb": 1.5,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Chat, Translation",
+      "ollamaCommand": "ollama run qwen2.5:1.5b-instruct-q4_K_M"
+    },
+    {
+      "model": "Qwen2.5 3B Instruct",
+      "family": "Qwen",
+      "params": 3,
+      "quantization": "Q4_K_M",
+      "minRamGb": 6,
+      "estimatedLoadGb": 2.5,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Chat, Coding",
+      "ollamaCommand": "ollama run qwen2.5:3b-instruct-q4_K_M"
+    },
+    {
+      "model": "Qwen2.5 7B Instruct",
+      "family": "Qwen",
+      "params": 7,
+      "quantization": "Q4_K_M",
+      "minRamGb": 10,
+      "estimatedLoadGb": 5.5,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Chat, Coding",
+      "ollamaCommand": "ollama run qwen2.5:7b-instruct-q4_K_M"
+    },
+    {
+      "model": "Qwen2.5 14B Instruct",
+      "family": "Qwen",
+      "params": 14,
+      "quantization": "Q4_K_M",
+      "minRamGb": 20,
+      "estimatedLoadGb": 11,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Coding, Chat",
+      "ollamaCommand": "ollama run qwen2.5:14b-instruct-q4_K_M"
+    },
+    {
+      "model": "Llama 3.2 3B Instruct",
+      "family": "Llama",
+      "params": 3,
+      "quantization": "Q4_K_M",
+      "minRamGb": 6,
+      "estimatedLoadGb": 2.5,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Chat",
+      "ollamaCommand": "ollama run llama3.2:3b-instruct-q4_K_M"
+    },
+    {
+      "model": "Llama 3.1 8B Instruct",
+      "family": "Llama",
+      "params": 8,
+      "quantization": "Q4_K_M",
+      "minRamGb": 12,
+      "estimatedLoadGb": 6.5,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Chat, Coding",
+      "ollamaCommand": "ollama run llama3.1:8b-instruct-q4_K_M"
+    },
+    {
+      "model": "Llama 3.1 8B Instruct (Q5)",
+      "family": "Llama",
+      "params": 8,
+      "quantization": "Q5_K_M",
+      "minRamGb": 14,
+      "estimatedLoadGb": 8,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Chat, Coding",
+      "ollamaCommand": "ollama run llama3.1:8b-instruct-q5_K_M"
+    },
+    {
+      "model": "Llama 3.1 70B Instruct",
+      "family": "Llama",
+      "params": 70,
+      "quantization": "Q4_K_M",
+      "minRamGb": 48,
+      "estimatedLoadGb": 42,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Quality, Coding",
+      "ollamaCommand": "ollama run llama3.1:70b-instruct-q4_K_M"
+    },
+    {
+      "model": "Mistral 7B Instruct",
+      "family": "Mistral",
+      "params": 7,
+      "quantization": "Q4_K_M",
+      "minRamGb": 10,
+      "estimatedLoadGb": 5.5,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Chat, Coding",
+      "ollamaCommand": "ollama run mistral:7b-instruct-q4_K_M"
+    },
+    {
+      "model": "Mixtral 8x7B Instruct",
+      "family": "Mistral",
+      "params": 46.7,
+      "quantization": "Q4_K_M",
+      "minRamGb": 36,
+      "estimatedLoadGb": 30,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Coding, Quality",
+      "ollamaCommand": "ollama run mixtral:8x7b"
+    },
+    {
+      "model": "Mistral Nemo 12B",
+      "family": "Mistral",
+      "params": 12,
+      "quantization": "Q4_K_M",
+      "minRamGb": 18,
+      "estimatedLoadGb": 9.5,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Chat, Translation",
+      "ollamaCommand": "ollama run mistral-nemo:12b"
+    },
+    {
+      "model": "Gemma 2 2B Instruct",
+      "family": "Gemma",
+      "params": 2,
+      "quantization": "Q4_K_M",
+      "minRamGb": 5,
+      "estimatedLoadGb": 1.8,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Chat",
+      "ollamaCommand": "ollama run gemma2:2b-instruct-q4_K_M"
+    },
+    {
+      "model": "Gemma 2 9B Instruct",
+      "family": "Gemma",
+      "params": 9,
+      "quantization": "Q4_K_M",
+      "minRamGb": 14,
+      "estimatedLoadGb": 7,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Chat, Coding",
+      "ollamaCommand": "ollama run gemma2:9b-instruct-q4_K_M"
+    },
+    {
+      "model": "Gemma 2 27B Instruct",
+      "family": "Gemma",
+      "params": 27,
+      "quantization": "Q4_K_M",
+      "minRamGb": 28,
+      "estimatedLoadGb": 21,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Quality, Coding",
+      "ollamaCommand": "ollama run gemma2:27b-instruct-q4_K_M"
+    },
+    {
+      "model": "Phi-3 Mini 3.8B",
+      "family": "Phi",
+      "params": 3.8,
+      "quantization": "Q4_K_M",
+      "minRamGb": 7,
+      "estimatedLoadGb": 3.2,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Coding, Chat",
+      "ollamaCommand": "ollama run phi3:mini"
+    },
+    {
+      "model": "Phi-3 Medium 14B",
+      "family": "Phi",
+      "params": 14,
+      "quantization": "Q4_K_M",
+      "minRamGb": 20,
+      "estimatedLoadGb": 11,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Coding, Quality",
+      "ollamaCommand": "ollama run phi3:medium"
+    },
+    {
+      "model": "Phi-4 14B",
+      "family": "Phi",
+      "params": 14,
+      "quantization": "Q4_K_M",
+      "minRamGb": 22,
+      "estimatedLoadGb": 11.5,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Coding, Quality",
+      "ollamaCommand": "ollama run phi4:14b-q4_K_M"
+    },
+    {
+      "model": "Qwen2.5 Coder 7B",
+      "family": "Qwen",
+      "params": 7,
+      "quantization": "Q4_K_M",
+      "minRamGb": 10,
+      "estimatedLoadGb": 5.5,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Coding",
+      "ollamaCommand": "ollama run qwen2.5-coder:7b"
+    },
+    {
+      "model": "Qwen2.5 Coder 14B",
+      "family": "Qwen",
+      "params": 14,
+      "quantization": "Q4_K_M",
+      "minRamGb": 22,
+      "estimatedLoadGb": 11,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Coding",
+      "ollamaCommand": "ollama run qwen2.5-coder:14b"
+    },
+    {
+      "model": "Llama 3.2 1B Instruct",
+      "family": "Llama",
+      "params": 1,
+      "quantization": "Q4_K_M",
+      "minRamGb": 3,
+      "estimatedLoadGb": 1,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Chat",
+      "ollamaCommand": "ollama run llama3.2:1b-instruct-q4_K_M"
+    },
+    {
+      "model": "Mistral Small 22B",
+      "family": "Mistral",
+      "params": 22,
+      "quantization": "Q4_K_M",
+      "minRamGb": 26,
+      "estimatedLoadGb": 17,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Coding, Quality",
+      "ollamaCommand": "ollama run mistral-small:22b"
+    },
+    {
+      "model": "Kimi K2 Instruct",
+      "family": "Kimi",
+      "params": 1000,
+      "quantization": "API",
+      "minRamGb": 0,
+      "estimatedLoadGb": 0,
+      "runsLocally": false,
+      "runtimes": [],
+      "bestFor": "Quality, Coding",
+      "ollamaCommand": null
+    },
+    {
+      "model": "Claude 3.5 Sonnet",
+      "family": "Claude",
+      "params": null,
+      "quantization": "API",
+      "minRamGb": 0,
+      "estimatedLoadGb": 0,
+      "runsLocally": false,
+      "runtimes": [],
+      "bestFor": "Coding, Chat",
+      "ollamaCommand": null
+    },
+    {
+      "model": "Claude 3.7 Sonnet",
+      "family": "Claude",
+      "params": null,
+      "quantization": "API",
+      "minRamGb": 0,
+      "estimatedLoadGb": 0,
+      "runsLocally": false,
+      "runtimes": [],
+      "bestFor": "Coding, Quality",
+      "ollamaCommand": null
+    },
+    {
+      "model": "Claude 3 Opus",
+      "family": "Claude",
+      "params": null,
+      "quantization": "API",
+      "minRamGb": 0,
+      "estimatedLoadGb": 0,
+      "runsLocally": false,
+      "runtimes": [],
+      "bestFor": "Quality, Analysis",
+      "ollamaCommand": null
+    },
+    {
+      "model": "Claude 4 Opus",
+      "family": "Claude",
+      "params": null,
+      "quantization": "API",
+      "minRamGb": 0,
+      "estimatedLoadGb": 0,
+      "runsLocally": false,
+      "runtimes": [],
+      "bestFor": "Quality, Coding",
+      "ollamaCommand": null
+    },
+    {
+      "model": "Qwen2.5 0.5B Instruct",
+      "family": "Qwen",
+      "params": 0.5,
+      "quantization": "Q4_K_M",
+      "minRamGb": 2,
+      "estimatedLoadGb": 0.8,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Chat, Mobile",
+      "ollamaCommand": "ollama run qwen2.5:0.5b-instruct-q4_K_M"
+    },
+    {
+      "model": "Gemma 3 1B Instruct",
+      "family": "Gemma",
+      "params": 1,
+      "quantization": "Q4_K_M",
+      "minRamGb": 3,
+      "estimatedLoadGb": 1,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Chat, Mobile",
+      "ollamaCommand": "ollama run gemma3:1b"
+    },
+    {
+      "model": "Phi-4 Mini 3.8B",
+      "family": "Phi",
+      "params": 3.8,
+      "quantization": "Q4_K_M",
+      "minRamGb": 7,
+      "estimatedLoadGb": 3.2,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Coding, Chat",
+      "ollamaCommand": "ollama run phi4-mini:3.8b"
+    },
+    {
+      "model": "SmolLM2 360M",
+      "family": "SmolLM",
+      "params": 0.36,
+      "quantization": "Q4_K_M",
+      "minRamGb": 1,
+      "estimatedLoadGb": 0.5,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Chat, Embedded",
+      "ollamaCommand": "ollama run smollm2:360m"
+    },
+    {
+      "model": "Llama 3.3 70B Instruct",
+      "family": "Llama",
+      "params": 70,
+      "quantization": "Q4_K_M",
+      "minRamGb": 48,
+      "estimatedLoadGb": 42,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Quality, Coding",
+      "ollamaCommand": "ollama run llama3.3:70b-instruct-q4_K_M"
+    },
+    {
+      "model": "Llama 3.1 405B Instruct",
+      "family": "Llama",
+      "params": 405,
+      "quantization": "Q4_K_M",
+      "minRamGb": 256,
+      "estimatedLoadGb": 243,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Quality, Reasoning, Coding",
+      "ollamaCommand": "ollama run llama3.1:405b-instruct-q4_K_M"
+    },
+    {
+      "model": "DeepSeek-R1 671B",
+      "family": "DeepSeek",
+      "params": 671,
+      "quantization": "Q4_K_M",
+      "minRamGb": 400,
+      "estimatedLoadGb": 380,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Reasoning, Coding",
+      "ollamaCommand": "ollama run deepseek-r1:671b-q4_K_M"
+    },
+    {
+      "model": "Qwen3 8B",
+      "family": "Qwen",
+      "params": 8,
+      "quantization": "Q4_K_M",
+      "minRamGb": 12,
+      "estimatedLoadGb": 6.5,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Chat, Coding",
+      "ollamaCommand": "ollama run qwen3:8b-q4_K_M"
+    },
+    {
+      "model": "Qwen3 14B",
+      "family": "Qwen",
+      "params": 14,
+      "quantization": "Q4_K_M",
+      "minRamGb": 20,
+      "estimatedLoadGb": 11,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Coding, Quality",
+      "ollamaCommand": "ollama run qwen3:14b-q4_K_M"
+    },
+    {
+      "model": "Qwen3 30B",
+      "family": "Qwen",
+      "params": 30,
+      "quantization": "Q4_K_M",
+      "minRamGb": 28,
+      "estimatedLoadGb": 22,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Quality, Coding",
+      "ollamaCommand": "ollama run qwen3:30b"
+    },
+    {
+      "model": "Gemma 3 4B Instruct",
+      "family": "Gemma",
+      "params": 4,
+      "quantization": "Q4_K_M",
+      "minRamGb": 8,
+      "estimatedLoadGb": 3.5,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Chat, Coding",
+      "ollamaCommand": "ollama run gemma3:4b"
+    },
+    {
+      "model": "Gemma 3 12B Instruct",
+      "family": "Gemma",
+      "params": 12,
+      "quantization": "Q4_K_M",
+      "minRamGb": 18,
+      "estimatedLoadGb": 9.5,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Chat, Quality",
+      "ollamaCommand": "ollama run gemma3:12b"
+    },
+    {
+      "model": "Gemma 3 27B Instruct",
+      "family": "Gemma",
+      "params": 27,
+      "quantization": "Q4_K_M",
+      "minRamGb": 28,
+      "estimatedLoadGb": 21,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Quality, Coding",
+      "ollamaCommand": "ollama run gemma3:27b"
+    },
+    {
+      "model": "DeepSeek-R1 Distill Qwen 7B",
+      "family": "DeepSeek",
+      "params": 7,
+      "quantization": "Q4_K_M",
+      "minRamGb": 10,
+      "estimatedLoadGb": 5.5,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Reasoning, Coding",
+      "ollamaCommand": "ollama run deepseek-r1:7b"
+    },
+    {
+      "model": "DeepSeek-R1 Distill Qwen 14B",
+      "family": "DeepSeek",
+      "params": 14,
+      "quantization": "Q4_K_M",
+      "minRamGb": 22,
+      "estimatedLoadGb": 11,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Reasoning, Quality",
+      "ollamaCommand": "ollama run deepseek-r1:14b"
+    },
+    {
+      "model": "DeepSeek-R1 Distill Llama 70B",
+      "family": "DeepSeek",
+      "params": 70,
+      "quantization": "Q4_K_M",
+      "minRamGb": 48,
+      "estimatedLoadGb": 42,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Reasoning, Quality",
+      "ollamaCommand": "ollama run deepseek-r1:70b"
+    },
+    {
+      "model": "GPT-4o",
+      "family": "OpenAI",
+      "params": null,
+      "quantization": "API",
+      "minRamGb": 0,
+      "estimatedLoadGb": 0,
+      "runsLocally": false,
+      "runtimes": [],
+      "bestFor": "Quality, Coding",
+      "ollamaCommand": null
+    },
+    {
+      "model": "GPT-4o mini",
+      "family": "OpenAI",
+      "params": null,
+      "quantization": "API",
+      "minRamGb": 0,
+      "estimatedLoadGb": 0,
+      "runsLocally": false,
+      "runtimes": [],
+      "bestFor": "Chat, Speed",
+      "ollamaCommand": null
+    },
+    {
+      "model": "Claude 4 Sonnet",
+      "family": "Claude",
+      "params": null,
+      "quantization": "API",
+      "minRamGb": 0,
+      "estimatedLoadGb": 0,
+      "runsLocally": false,
+      "runtimes": [],
+      "bestFor": "Coding, Quality",
+      "ollamaCommand": null
+    },
+    {
+      "model": "Gemini 2.5 Pro",
+      "family": "Google",
+      "params": null,
+      "quantization": "API",
+      "minRamGb": 0,
+      "estimatedLoadGb": 0,
+      "runsLocally": false,
+      "runtimes": [],
+      "bestFor": "Quality, Analysis",
+      "ollamaCommand": null
+    },
+    {
+      "model": "Gemini 2.5 Flash",
+      "family": "Google",
+      "params": null,
+      "quantization": "API",
+      "minRamGb": 0,
+      "estimatedLoadGb": 0,
+      "runsLocally": false,
+      "runtimes": [],
+      "bestFor": "Speed, Chat",
+      "ollamaCommand": null
+    },
+    {
+      "model": "DeepSeek-V3",
+      "family": "DeepSeek",
+      "params": 671,
+      "quantization": "API",
+      "minRamGb": 0,
+      "estimatedLoadGb": 0,
+      "runsLocally": false,
+      "runtimes": [],
+      "bestFor": "Quality, Coding",
+      "ollamaCommand": null
+    },
+    {
+      "model": "GLM-5",
+      "family": "Zhipu",
+      "params": 744,
+      "quantization": "API",
+      "minRamGb": 0,
+      "estimatedLoadGb": 0,
+      "runsLocally": false,
+      "runtimes": [],
+      "bestFor": "Reasoning, Quality",
+      "ollamaCommand": null
+    },
+    {
+      "model": "GLM-4 Plus",
+      "family": "Zhipu",
+      "params": null,
+      "quantization": "API",
+      "minRamGb": 0,
+      "estimatedLoadGb": 0,
+      "runsLocally": false,
+      "runtimes": [],
+      "bestFor": "Quality, Analysis",
+      "ollamaCommand": null
+    },
+    {
+      "model": "Qwen3 235B A22B",
+      "family": "Qwen",
+      "params": 235,
+      "quantization": "Q4_K_M",
+      "minRamGb": 192,
+      "estimatedLoadGb": 130,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Quality, Reasoning",
+      "ollamaCommand": "ollama run qwen3:235b-a22b-q4_K_M"
+    },
+    {
+      "model": "Mistral Small 3.1",
+      "family": "Mistral",
+      "params": 24,
+      "quantization": "Q4_K_M",
+      "minRamGb": 24,
+      "estimatedLoadGb": 15,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Chat, Coding",
+      "ollamaCommand": "ollama run mistral-small3.1:24b"
+    },
+    {
+      "model": "DeepSeek-V3-0324",
+      "family": "DeepSeek",
+      "params": 671,
+      "quantization": "API",
+      "minRamGb": 0,
+      "estimatedLoadGb": 0,
+      "runsLocally": false,
+      "runtimes": [],
+      "bestFor": "Quality, Coding",
+      "ollamaCommand": null
+    },
+    {
+      "model": "DeepSeek-R1",
+      "family": "DeepSeek",
+      "params": 671,
+      "quantization": "API",
+      "minRamGb": 0,
+      "estimatedLoadGb": 0,
+      "runsLocally": false,
+      "runtimes": [],
+      "bestFor": "Reasoning, Quality",
+      "ollamaCommand": null
+    },
+    {
+      "model": "Qwen3.5 0.8B Instruct",
+      "family": "Qwen",
+      "params": 0.8,
+      "quantization": "Q4_K_M",
+      "minRamGb": 2,
+      "estimatedLoadGb": 0.8,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Chat, Mobile",
+      "ollamaCommand": "ollama run qwen3.5:0.8b"
+    },
+    {
+      "model": "Qwen3.5 2B Instruct",
+      "family": "Qwen",
+      "params": 2,
+      "quantization": "Q4_K_M",
+      "minRamGb": 4,
+      "estimatedLoadGb": 1.8,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Chat, Edge tasks",
+      "ollamaCommand": "ollama run qwen3.5:2b"
+    },
+    {
+      "model": "Qwen3.5 4B Instruct",
+      "family": "Qwen",
+      "params": 4,
+      "quantization": "Q4_K_M",
+      "minRamGb": 8,
+      "estimatedLoadGb": 3.5,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Coding, Agents, Multimodal",
+      "ollamaCommand": "ollama run qwen3.5:4b"
+    },
+    {
+      "model": "Qwen3.5 9B Instruct",
+      "family": "Qwen",
+      "params": 9,
+      "quantization": "Q4_K_M",
+      "minRamGb": 14,
+      "estimatedLoadGb": 7,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Quality, Coding, Reasoning",
+      "ollamaCommand": "ollama run qwen3.5:9b"
+    },
+    {
+      "model": "Qwen3.5 35B-A3B Instruct",
+      "family": "Qwen",
+      "params": 35,
+      "quantization": "Q4_K_M",
+      "minRamGb": 24,
+      "estimatedLoadGb": 20,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Reasoning, Coding, Agent scenarios",
+      "ollamaCommand": "ollama run qwen3.5:35b-a3b"
+    },
+    {
+      "model": "Qwen3.5 27B Instruct",
+      "family": "Qwen",
+      "params": 27,
+      "quantization": "Q4_K_M",
+      "minRamGb": 20,
+      "estimatedLoadGb": 16,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Chat, Coding, Complex reasoning",
+      "ollamaCommand": "ollama run qwen3.5:27b"
+    },
+    {
+      "model": "Qwen3.5 122B-A10B Instruct",
+      "family": "Qwen",
+      "params": 122,
+      "quantization": "Q4_K_M",
+      "minRamGb": 96,
+      "estimatedLoadGb": 72,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Frontier-level reasoning, Complex tasks",
+      "ollamaCommand": "ollama run qwen3.5:122b-a10b"
+    },
+    {
+      "model": "LFM2 24B-A2B Instruct",
+      "family": "LFM2",
+      "params": 24,
+      "quantization": "Q4_K_M",
+      "minRamGb": 16,
+      "estimatedLoadGb": 14,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Local AI agents, privacy-first tool calling, MCP workflows",
+      "ollamaCommand": "ollama run lfm2:24b-a2b"
+    },
+    {
+      "model": "LFM2.5 8B-A1B",
+      "family": "LFM2",
+      "params": 8.3,
+      "quantization": "Q4_K_M",
+      "minRamGb": 10,
+      "estimatedLoadGb": 5.5,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "On-device agents, tool calling, multilingual chat",
+      "ollamaCommand": "ollama run lfm2.5:8b-a1b-q4_K_M"
+    },
+    {
+      "model": "Granite 4.1 3B Instruct",
+      "family": "Granite",
+      "params": 3,
+      "quantization": "Q4_K_M",
+      "minRamGb": 6,
+      "estimatedLoadGb": 2,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Lightweight chat, classification, edge tasks",
+      "ollamaCommand": "ollama run granite4.1:3b"
+    },
+    {
+      "model": "Granite 4.1 8B Instruct",
+      "family": "Granite",
+      "params": 8,
+      "quantization": "Q4_K_M",
+      "minRamGb": 12,
+      "estimatedLoadGb": 5.5,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Enterprise assistant, tool calling, instruction following",
+      "ollamaCommand": "ollama run granite4.1:8b"
+    },
+    {
+      "model": "Qwen3.6 27B",
+      "family": "Qwen",
+      "params": 27,
+      "quantization": "Q4_K_M",
+      "minRamGb": 24,
+      "estimatedLoadGb": 18,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Coding, Quality, Long context",
+      "ollamaCommand": "ollama run qwen3.6:27b"
+    },
+    {
+      "model": "Qwen3.6 35B-A3B",
+      "family": "Qwen",
+      "params": 35,
+      "quantization": "Q4_K_M",
+      "minRamGb": 24,
+      "estimatedLoadGb": 22,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Reasoning, Coding, Agents",
+      "ollamaCommand": "ollama run qwen3.6:35b-a3b"
+    },
+    {
+      "model": "Gemma 4 31B",
+      "family": "Gemma",
+      "params": 31,
+      "quantization": "Q4_K_M",
+      "minRamGb": 32,
+      "estimatedLoadGb": 20,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Quality, Coding, Multimodal",
+      "ollamaCommand": "ollama run gemma4:31b"
+    },
+    {
+      "model": "Gemma 4 26B-A4B",
+      "family": "Gemma",
+      "params": 26,
+      "quantization": "Q4_K_M",
+      "minRamGb": 24,
+      "estimatedLoadGb": 16,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Chat, Coding, Multimodal",
+      "ollamaCommand": "ollama run gemma4:26b"
+    },
+    {
+      "model": "Gemma 4 E4B",
+      "family": "Gemma",
+      "params": 4.5,
+      "quantization": "Q4_K_M",
+      "minRamGb": 8,
+      "estimatedLoadGb": 4,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "On-device, Mobile, Chat",
+      "ollamaCommand": "ollama run gemma4:e4b"
+    },
+    {
+      "model": "Gemma 4 E2B",
+      "family": "Gemma",
+      "params": 2.3,
+      "quantization": "Q4_K_M",
+      "minRamGb": 4,
+      "estimatedLoadGb": 2.3,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "IoT, Mobile, Edge",
+      "ollamaCommand": "ollama run gemma4:e2b"
+    },
+    {
+      "model": "Llama 4 Scout",
+      "family": "Llama",
+      "params": 109,
+      "quantization": "Q4_K_M",
+      "minRamGb": 80,
+      "estimatedLoadGb": 67,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Long context, Quality, Multimodal",
+      "ollamaCommand": "ollama run llama4:scout"
+    },
+    {
+      "model": "Llama 4 Maverick",
+      "family": "Llama",
+      "params": 400,
+      "quantization": "Q4_K_M",
+      "minRamGb": 256,
+      "estimatedLoadGb": 245,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Frontier quality, Long context",
+      "ollamaCommand": "ollama run llama4:maverick"
+    },
+    {
+      "model": "Mistral Medium 3.5",
+      "family": "Mistral",
+      "params": 128,
+      "quantization": "API",
+      "minRamGb": 0,
+      "estimatedLoadGb": 0,
+      "runsLocally": false,
+      "runtimes": [],
+      "bestFor": "Coding, Quality, Tool calling",
+      "ollamaCommand": null
+    },
+    {
+      "model": "DeepSeek V4 Flash",
+      "family": "DeepSeek",
+      "params": 284,
+      "quantization": "API",
+      "minRamGb": 0,
+      "estimatedLoadGb": 0,
+      "runsLocally": false,
+      "runtimes": [],
+      "bestFor": "Reasoning, Coding, Long context",
+      "ollamaCommand": null
+    },
+    {
+      "model": "DeepSeek V4 Pro",
+      "family": "DeepSeek",
+      "params": 1600,
+      "quantization": "API",
+      "minRamGb": 0,
+      "estimatedLoadGb": 0,
+      "runsLocally": false,
+      "runtimes": [],
+      "bestFor": "Frontier reasoning, Coding",
+      "ollamaCommand": null
+    },
+    {
+      "model": "Kimi K2.6",
+      "family": "Kimi",
+      "params": 1000,
+      "quantization": "API",
+      "minRamGb": 0,
+      "estimatedLoadGb": 0,
+      "runsLocally": false,
+      "runtimes": [],
+      "bestFor": "Coding, Agentic tasks",
+      "ollamaCommand": null
+    },
+    {
+      "model": "GLM-5.1",
+      "family": "Zhipu",
+      "params": 744,
+      "quantization": "API",
+      "minRamGb": 0,
+      "estimatedLoadGb": 0,
+      "runsLocally": false,
+      "runtimes": [],
+      "bestFor": "Coding, Reasoning, Long context",
+      "ollamaCommand": null
+    },
+    {
+      "model": "Claude Opus 4.7",
+      "family": "Claude",
+      "params": null,
+      "quantization": "API",
+      "minRamGb": 0,
+      "estimatedLoadGb": 0,
+      "runsLocally": false,
+      "runtimes": [],
+      "bestFor": "Coding, Quality, Reasoning",
+      "ollamaCommand": null
+    },
+    {
+      "model": "Claude Opus 4.8",
+      "family": "Claude",
+      "params": null,
+      "quantization": "API",
+      "minRamGb": 0,
+      "estimatedLoadGb": 0,
+      "runsLocally": false,
+      "runtimes": [],
+      "bestFor": "Coding, Agentic, Reasoning",
+      "ollamaCommand": null
+    },
+    {
+      "model": "GPT-5.5",
+      "family": "OpenAI",
+      "params": null,
+      "quantization": "API",
+      "minRamGb": 0,
+      "estimatedLoadGb": 0,
+      "runsLocally": false,
+      "runtimes": [],
+      "bestFor": "Quality, Reasoning, Multimodal",
+      "ollamaCommand": null
+    },
+    {
+      "model": "Gemini 3.1 Pro",
+      "family": "Google",
+      "params": null,
+      "quantization": "API",
+      "minRamGb": 0,
+      "estimatedLoadGb": 0,
+      "runsLocally": false,
+      "runtimes": [],
+      "bestFor": "Multimodal, Long context, Quality",
+      "ollamaCommand": null
+    },
+    {
+      "model": "Grok 4.3",
+      "family": "xAI",
+      "params": null,
+      "quantization": "API",
+      "minRamGb": 0,
+      "estimatedLoadGb": 0,
+      "runsLocally": false,
+      "runtimes": [],
+      "bestFor": "Reasoning, Real-time data, Coding",
+      "ollamaCommand": null
+    },
+    {
+      "model": "Gemma 4 12B",
+      "family": "Gemma",
+      "params": 12,
+      "quantization": "Q4_K_M",
+      "minRamGb": 16,
+      "estimatedLoadGb": 8,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Chat, Coding, Multimodal",
+      "ollamaCommand": "ollama run gemma4:12b"
+    },
+    {
+      "model": "Claude Fable 5",
+      "family": "Claude",
+      "params": null,
+      "quantization": "API",
+      "minRamGb": 0,
+      "estimatedLoadGb": 0,
+      "runsLocally": false,
+      "runtimes": [],
+      "bestFor": "Coding, Agentic, Reasoning",
+      "ollamaCommand": null
+    },
+    {
+      "model": "GLM-5.2",
+      "family": "Zhipu",
+      "params": null,
+      "quantization": "API",
+      "minRamGb": 0,
+      "estimatedLoadGb": 0,
+      "runsLocally": false,
+      "runtimes": [],
+      "bestFor": "Coding, Reasoning, Agentic",
+      "ollamaCommand": null
+    },
+    {
+      "model": "Kimi K2.7-Code",
+      "family": "Kimi",
+      "params": 1000,
+      "quantization": "API",
+      "minRamGb": 0,
+      "estimatedLoadGb": 0,
+      "runsLocally": false,
+      "runtimes": [],
+      "bestFor": "Coding, Agentic tasks",
+      "ollamaCommand": null
+    },
+    {
+      "model": "MiniMax M3",
+      "family": "MiniMax",
+      "params": 428,
+      "quantization": "API",
+      "minRamGb": 0,
+      "estimatedLoadGb": 0,
+      "runsLocally": false,
+      "runtimes": [],
+      "bestFor": "Coding, Agentic, Long context",
+      "ollamaCommand": null
+    },
+    {
+      "model": "Qwen3.7-Plus",
+      "family": "Qwen",
+      "params": null,
+      "quantization": "API",
+      "minRamGb": 0,
+      "estimatedLoadGb": 0,
+      "runsLocally": false,
+      "runtimes": [],
+      "bestFor": "Multimodal, Agentic, Vision",
+      "ollamaCommand": null
+    },
+    {
+      "model": "NVIDIA Nemotron 3 Ultra",
+      "family": "Nemotron",
+      "params": 550,
+      "quantization": "API",
+      "minRamGb": 0,
+      "estimatedLoadGb": 0,
+      "runsLocally": false,
+      "runtimes": [],
+      "bestFor": "Reasoning, Agentic, Long context",
+      "ollamaCommand": null
+    },
+    {
+      "model": "NVIDIA Nemotron Cascade 2 30B-A3B",
+      "family": "Nemotron",
+      "params": 30,
+      "quantization": "Q6_K",
+      "minRamGb": 32,
+      "estimatedLoadGb": 24,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Reasoning, Math, Agentic tasks",
+      "ollamaCommand": "ollama run nemotron-cascade-2:30b"
+    },
+    {
+      "model": "Poolside Laguna XS.2",
+      "family": "Laguna",
+      "params": 33,
+      "quantization": "Q4_K_M",
+      "minRamGb": 32,
+      "estimatedLoadGb": 23,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Agentic coding, long-horizon software engineering",
+      "ollamaCommand": "ollama run laguna-xs.2:q4_K_M"
+    },
+    {
+      "model": "Cohere North Mini Code",
+      "family": "North",
+      "params": 30,
+      "quantization": "Q4_K_M",
+      "minRamGb": 24,
+      "estimatedLoadGb": 19,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Agentic coding, repo-scale software engineering",
+      "ollamaCommand": "ollama run north-mini-code-1.0:q4_K_M"
+    },
+    {
+      "model": "GPT-OSS 120B",
+      "family": "GPT-OSS",
+      "params": 117,
+      "quantization": "MXFP4",
+      "minRamGb": 96,
+      "estimatedLoadGb": 65.4,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Reasoning, Coding, Agents",
+      "ollamaCommand": "ollama run gpt-oss:120b"
+    },
+    {
+      "model": "GPT-OSS 20B",
+      "family": "GPT-OSS",
+      "params": 21,
+      "quantization": "MXFP4",
+      "minRamGb": 24,
+      "estimatedLoadGb": 13.8,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Chat, Coding, Reasoning",
+      "ollamaCommand": "ollama run gpt-oss:20b"
+    },
+    {
+      "model": "Qwen3-Next 80B-A3B",
+      "family": "Qwen",
+      "params": 80,
+      "quantization": "Q4_K_M",
+      "minRamGb": 72,
+      "estimatedLoadGb": 50.4,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Chat, Coding, Long Context",
+      "ollamaCommand": "ollama run qwen3-next:80b"
+    },
+    {
+      "model": "Qwen3.6 35B-A3B (Q8)",
+      "family": "Qwen",
+      "params": 35,
+      "quantization": "Q8_0",
+      "minRamGb": 64,
+      "estimatedLoadGb": 38.7,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Reasoning, Coding, Agents",
+      "ollamaCommand": "ollama run qwen3.6:35b-a3b-q8_0"
+    },
+    {
+      "model": "Qwen3.5 35B-A3B Instruct (Q8)",
+      "family": "Qwen",
+      "params": 35,
+      "quantization": "Q8_0",
+      "minRamGb": 64,
+      "estimatedLoadGb": 38.7,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Reasoning, Coding, Agent scenarios",
+      "ollamaCommand": "ollama run qwen3.5:35b-a3b-q8_0"
+    },
+    {
+      "model": "Qwen3.5 9B Instruct (Q8)",
+      "family": "Qwen",
+      "params": 9,
+      "quantization": "Q8_0",
+      "minRamGb": 16,
+      "estimatedLoadGb": 10.7,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Quality, Coding, Reasoning",
+      "ollamaCommand": "ollama run qwen3.5:9b-q8_0"
+    },
+    {
+      "model": "Qwen3.6 27B (Q8)",
+      "family": "Qwen",
+      "params": 27,
+      "quantization": "Q8_0",
+      "minRamGb": 48,
+      "estimatedLoadGb": 30,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Coding, Quality, Long context",
+      "ollamaCommand": "ollama run qwen3.6:27b-q8_0"
+    },
+    {
+      "model": "Gemma 4 12B (Q8)",
+      "family": "Gemma",
+      "params": 12,
+      "quantization": "Q8_0",
+      "minRamGb": 24,
+      "estimatedLoadGb": 12.8,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Chat, Coding, Multimodal",
+      "ollamaCommand": "ollama run gemma4:12b-it-q8_0"
+    },
+    {
+      "model": "Gemma 4 26B-A4B (Q8)",
+      "family": "Gemma",
+      "params": 26,
+      "quantization": "Q8_0",
+      "minRamGb": 48,
+      "estimatedLoadGb": 28.1,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Chat, Coding, Multimodal",
+      "ollamaCommand": "ollama run gemma4:26b-a4b-it-q8_0"
+    },
+    {
+      "model": "Qwen3-Next 80B-A3B (Q8)",
+      "family": "Qwen",
+      "params": 80,
+      "quantization": "Q8_0",
+      "minRamGb": 128,
+      "estimatedLoadGb": 84.8,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Chat, Coding, Long Context",
+      "ollamaCommand": "ollama run qwen3-next:80b-a3b-instruct-q8_0"
+    },
+    {
+      "model": "Qwen3 14B (Q8)",
+      "family": "Qwen",
+      "params": 14,
+      "quantization": "Q8_0",
+      "minRamGb": 24,
+      "estimatedLoadGb": 15.9,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Coding, Quality",
+      "ollamaCommand": "ollama run qwen3:14b-q8_0"
+    },
+    {
+      "model": "Llama 3.3 70B Instruct (Q8)",
+      "family": "Llama",
+      "params": 70,
+      "quantization": "Q8_0",
+      "minRamGb": 128,
+      "estimatedLoadGb": 75,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Quality, Coding",
+      "ollamaCommand": "ollama run llama3.3:70b-instruct-q8_0"
+    },
+    {
+      "model": "Llama 3.3 70B Instruct (Q6)",
+      "family": "Llama",
+      "params": 70,
+      "quantization": "Q6_K",
+      "minRamGb": 96,
+      "estimatedLoadGb": 57.9,
+      "runsLocally": true,
+      "runtimes": [
+        "ollama",
+        "llama.cpp",
+        "lm-studio"
+      ],
+      "bestFor": "Quality, Coding",
+      "ollamaCommand": "ollama run llama3.3:70b-instruct-q6_K"
+    }
+  ]
+}