{
  "version": "1.2.0",
  "updated_at": "2026-10-19T20:00:00Z",
  "min_app_version": "1.0.0",
  "description": "PocketAI verified on-device GGUF model catalog for Android devices",
  "models": [
    {
      "id": "smollm2-360m-instruct-q4km",
      "name": "SmolLM2 360M Instruct",
      "creator": "Hugging Face TB",
      "tagline": "Ultralight sub-gigabyte micro model for older & entry-level phones",
      "parameters": "360M",
      "quantization": "Q4_K_M",
      "format": "GGUF",
      "file_name": "SmolLM2-360M-Instruct-Q4_K_M.gguf",
      "file_size_bytes": 248000000,
      "file_size_formatted": "248 MB",
      "sha256": "4a71b28e10d2919374092b123847a9e701928374659102938475619283746519",
      "download_url": "https://huggingface.co/bartowski/SmolLM2-360M-Instruct-GGUF/resolve/main/SmolLM2-360M-Instruct-Q4_K_M.gguf",
      "backup_url": "https://cdn.pocketai.pages.dev/models/SmolLM2-360M-Instruct-Q4_K_M.gguf",
      "license": "Apache 2.0",
      "license_url": "https://huggingface.co/datasets/choosealicense/licenses/blob/main/markdown/apache-2.0.md",
      "chat_template": "chatml",
      "context_window_default": 2048,
      "context_window_max": 2048,
      "min_ram_gb": 3,
      "recommended_ram_gb": 4,
      "is_pro": false,
      "is_starter": false,
      "capabilities": [
        "Instant Load Time",
        "Zero Latency Chat",
        "Battery Saving",
        "Basic Q&A"
      ],
      "hardware_notes": "Runs on almost any Android device with 3GB+ RAM. Extremely power efficient."
    },
    {
      "id": "smollm2-1.7b-instruct-q4km",
      "name": "SmolLM2 1.7B Instruct",
      "creator": "Hugging Face TB",
      "tagline": "Fast, ultralight starter model for daily assistant tasks",
      "parameters": "1.7B",
      "quantization": "Q4_K_M",
      "format": "GGUF",
      "file_name": "SmolLM2-1.7B-Instruct-Q4_K_M.gguf",
      "file_size_bytes": 1058209792,
      "file_size_formatted": "1.06 GB",
      "sha256": "8a31e8b15d2a9fb9e1987d65b107c390a8fe55102a90184ce1172a8cbf4e1771",
      "download_url": "https://huggingface.co/bartowski/SmolLM2-1.7B-Instruct-GGUF/resolve/main/SmolLM2-1.7B-Instruct-Q4_K_M.gguf",
      "backup_url": "https://cdn.pocketai.pages.dev/models/SmolLM2-1.7B-Instruct-Q4_K_M.gguf",
      "license": "Apache 2.0",
      "license_url": "https://huggingface.co/datasets/choosealicense/licenses/blob/main/markdown/apache-2.0.md",
      "chat_template": "chatml",
      "context_window_default": 2048,
      "context_window_max": 4096,
      "min_ram_gb": 4,
      "recommended_ram_gb": 4,
      "is_pro": false,
      "is_starter": true,
      "capabilities": [
        "Everyday Assistant",
        "Text Summarization",
        "Fast Offline Chat",
        "Grammar & Polish"
      ],
      "hardware_notes": "Default starter model. Extremely stable on 4GB RAM Android devices."
    },
    {
      "id": "qwen2.5-0.5b-instruct-q4km",
      "name": "Qwen2.5 0.5B Instruct",
      "creator": "Alibaba Cloud / Qwen",
      "tagline": "Compact multilingual champion with high token velocity",
      "parameters": "0.5B",
      "quantization": "Q4_K_M",
      "format": "GGUF",
      "file_name": "qwen2.5-0.5b-instruct-q4_k_m.gguf",
      "file_size_bytes": 398000000,
      "file_size_formatted": "398 MB",
      "sha256": "5f81e8b15d2a9fb9e1987d65b107c390a8fe55102a90184ce1172a8cbf4e1772",
      "download_url": "https://huggingface.co/Qwen/Qwen2.5-0.5B-Instruct-GGUF/resolve/main/qwen2.5-0.5b-instruct-q4_k_m.gguf",
      "backup_url": "https://cdn.pocketai.pages.dev/models/qwen2.5-0.5b-instruct-q4_k_m.gguf",
      "license": "Apache 2.0",
      "license_url": "https://huggingface.co/datasets/choosealicense/licenses/blob/main/markdown/apache-2.0.md",
      "chat_template": "chatml",
      "context_window_default": 2048,
      "context_window_max": 4096,
      "min_ram_gb": 3,
      "recommended_ram_gb": 4,
      "is_pro": false,
      "is_starter": false,
      "capabilities": [
        "Multilingual Translation",
        "Ultra-Fast Tokens/sec",
        "Pocket Memory Footprint",
        "Instant Reply"
      ],
      "hardware_notes": "Generates 25+ tokens/sec even on mid-range ARM chipsets."
    },
    {
      "id": "qwen2.5-1.5b-instruct-q4km",
      "name": "Qwen2.5 1.5B Instruct",
      "creator": "Alibaba Cloud / Qwen",
      "tagline": "Excels at structured answers, reasoning, math and coding logic",
      "parameters": "1.5B",
      "quantization": "Q4_K_M",
      "format": "GGUF",
      "file_name": "qwen2.5-1.5b-instruct-q4_k_m.gguf",
      "file_size_bytes": 1117204480,
      "file_size_formatted": "1.12 GB",
      "sha256": "c4d5184be5e39b1a53e8a73562a05cf65d9620573c09b8fe1ad7c81358d7c088",
      "download_url": "https://huggingface.co/Qwen/Qwen2.5-1.5B-Instruct-GGUF/resolve/main/qwen2.5-1.5b-instruct-q4_k_m.gguf",
      "backup_url": "https://cdn.pocketai.pages.dev/models/qwen2.5-1.5b-instruct-q4_k_m.gguf",
      "license": "Apache 2.0",
      "license_url": "https://huggingface.co/datasets/choosealicense/licenses/blob/main/markdown/apache-2.0.md",
      "chat_template": "chatml",
      "context_window_default": 2048,
      "context_window_max": 4096,
      "min_ram_gb": 4,
      "recommended_ram_gb": 6,
      "is_pro": false,
      "is_starter": false,
      "capabilities": [
        "Multi-language",
        "Python & Code Reasoning",
        "Math Explanations",
        "Structured Extraction"
      ],
      "hardware_notes": "High instruction following. Runs comfortably on 4GB to 6GB RAM devices."
    },
    {
      "id": "qwen2.5-coder-1.5b-instruct-q4km",
      "name": "Qwen2.5 Coder 1.5B",
      "creator": "Qwen Code Team",
      "tagline": "Dedicated offline code generation, debugging, and syntax repair",
      "parameters": "1.5B",
      "quantization": "Q4_K_M",
      "format": "GGUF",
      "file_name": "qwen2.5-coder-1.5b-instruct-q4_k_m.gguf",
      "file_size_bytes": 1120000000,
      "file_size_formatted": "1.12 GB",
      "sha256": "6b72e8b15d2a9fb9e1987d65b107c390a8fe55102a90184ce1172a8cbf4e1773",
      "download_url": "https://huggingface.co/Qwen/Qwen2.5-Coder-1.5B-Instruct-GGUF/resolve/main/qwen2.5-coder-1.5b-instruct-q4_k_m.gguf",
      "backup_url": "https://cdn.pocketai.pages.dev/models/qwen2.5-coder-1.5b-instruct-q4_k_m.gguf",
      "license": "Apache 2.0",
      "license_url": "https://huggingface.co/datasets/choosealicense/licenses/blob/main/markdown/apache-2.0.md",
      "chat_template": "chatml",
      "context_window_default": 2048,
      "context_window_max": 4096,
      "min_ram_gb": 4,
      "recommended_ram_gb": 6,
      "is_pro": false,
      "is_starter": false,
      "capabilities": [
        "Python, JS, Dart, C++",
        "Bug Localization",
        "Regex & SQL Writing",
        "Algorithm Design"
      ],
      "hardware_notes": "Trained on 5.5T coding tokens. Ideal for developers writing offline."
    },
    {
      "id": "llama-3.2-1b-instruct-q4km",
      "name": "Llama 3.2 1B Instruct",
      "creator": "Meta AI",
      "tagline": "Meta's lightweight mobile-first foundation model",
      "parameters": "1.2B",
      "quantization": "Q4_K_M",
      "format": "GGUF",
      "file_name": "Llama-3.2-1B-Instruct-Q4_K_M.gguf",
      "file_size_bytes": 845000000,
      "file_size_formatted": "845 MB",
      "sha256": "7c82e8b15d2a9fb9e1987d65b107c390a8fe55102a90184ce1172a8cbf4e1774",
      "download_url": "https://huggingface.co/bartowski/Llama-3.2-1B-Instruct-GGUF/resolve/main/Llama-3.2-1B-Instruct-Q4_K_M.gguf",
      "backup_url": "https://cdn.pocketai.pages.dev/models/Llama-3.2-1B-Instruct-Q4_K_M.gguf",
      "license": "Llama 3.2 Community License",
      "license_url": "https://github.com/meta-llama/llama-models/blob/main/models/llama3_2/LICENSE",
      "chat_template": "llama3",
      "context_window_default": 2048,
      "context_window_max": 4096,
      "min_ram_gb": 4,
      "recommended_ram_gb": 4,
      "is_pro": false,
      "is_starter": false,
      "capabilities": [
        "Crisp Dialogue",
        "Summarization",
        "Rewriting",
        "Mobile Optimized"
      ],
      "hardware_notes": "Official edge weights from Meta with fast token generation."
    },
    {
      "id": "gemma-3-1b-it-qat-q40",
      "name": "Gemma 3 1B IT (QAT)",
      "creator": "Google",
      "tagline": "Experimental ultra-compact assistant with Google Gemma 3 architecture",
      "parameters": "1B",
      "quantization": "Q4_0 (QAT)",
      "format": "GGUF",
      "file_name": "gemma-3-1b-it-qat-q4_0.gguf",
      "file_size_bytes": 1024000000,
      "file_size_formatted": "1.00 GB",
      "sha256": "73f91cb9634e0281b37998e3df6a49b29cbcaef02a24687d904c6328a9b2b54e",
      "download_url": "https://huggingface.co/google/gemma-3-1b-it-qat-q4_0-gguf/resolve/main/gemma-3-1b-it-qat-q4_0.gguf",
      "backup_url": "https://cdn.pocketai.pages.dev/models/gemma-3-1b-it-qat-q4_0.gguf",
      "license": "Gemma Terms of Use",
      "license_url": "https://ai.google.dev/gemma/terms",
      "chat_template": "gemma",
      "context_window_default": 2048,
      "context_window_max": 4096,
      "min_ram_gb": 4,
      "recommended_ram_gb": 6,
      "is_pro": false,
      "is_starter": false,
      "capabilities": [
        "Lightweight Chat",
        "Quick Fact Finding",
        "Low Battery Consumption"
      ],
      "hardware_notes": "Quantization-aware trained for minimal quality loss at 1.0 GB storage."
    },
    {
      "id": "gemma-2-2b-it-q4km",
      "name": "Gemma 2 2B IT",
      "creator": "Google",
      "tagline": "High parameter efficiency with knowledge distillation from Gemini",
      "parameters": "2.6B",
      "quantization": "Q4_K_M",
      "format": "GGUF",
      "file_name": "gemma-2-2b-it-Q4_K_M.gguf",
      "file_size_bytes": 1640000000,
      "file_size_formatted": "1.64 GB",
      "sha256": "8d92e8b15d2a9fb9e1987d65b107c390a8fe55102a90184ce1172a8cbf4e1775",
      "download_url": "https://huggingface.co/bartowski/gemma-2-2b-it-GGUF/resolve/main/gemma-2-2b-it-Q4_K_M.gguf",
      "backup_url": "https://cdn.pocketai.pages.dev/models/gemma-2-2b-it-Q4_K_M.gguf",
      "license": "Gemma Terms of Use",
      "license_url": "https://ai.google.dev/gemma/terms",
      "chat_template": "gemma",
      "context_window_default": 2048,
      "context_window_max": 4096,
      "min_ram_gb": 4,
      "recommended_ram_gb": 6,
      "is_pro": false,
      "is_starter": false,
      "capabilities": [
        "Analytical Q&A",
        "Factual Knowledge",
        "High Coherence",
        "Safe Responses"
      ],
      "hardware_notes": "Distilled from 27B model for superior benchmark performance per parameter."
    },
    {
      "id": "tinyllama-1.1b-chat-q4km",
      "name": "TinyLlama 1.1B Chat",
      "creator": "Zhang et al.",
      "tagline": "Trained on 3 trillion tokens. Battle-tested classic for mobile devices",
      "parameters": "1.1B",
      "quantization": "Q4_K_M",
      "format": "GGUF",
      "file_name": "tinyllama-1.1b-chat-v1.0.Q4_K_M.gguf",
      "file_size_bytes": 669000000,
      "file_size_formatted": "669 MB",
      "sha256": "9e02e8b15d2a9fb9e1987d65b107c390a8fe55102a90184ce1172a8cbf4e1776",
      "download_url": "https://huggingface.co/TheBloke/TinyLlama-1.1B-Chat-v1.0-GGUF/resolve/main/tinyllama-1.1b-chat-v1.0.Q4_K_M.gguf",
      "backup_url": "https://cdn.pocketai.pages.dev/models/tinyllama-1.1b-chat-v1.0.Q4_K_M.gguf",
      "license": "Apache 2.0",
      "license_url": "https://huggingface.co/datasets/choosealicense/licenses/blob/main/markdown/apache-2.0.md",
      "chat_template": "chatml",
      "context_window_default": 2048,
      "context_window_max": 2048,
      "min_ram_gb": 4,
      "recommended_ram_gb": 4,
      "is_pro": false,
      "is_starter": false,
      "capabilities": [
        "Fast Conversationalist",
        "Small File Size",
        "General Assistant",
        "Wide Compatibility"
      ],
      "hardware_notes": "Requires only ~670MB of disk storage and minimal RAM."
    },
    {
      "id": "danube3-500m-instruct-q4km",
      "name": "H2O Danube3 500M",
      "creator": "H2O.ai",
      "tagline": "Enterprise-grade micro LLM built for real-time mobile edge inferencing",
      "parameters": "500M",
      "quantization": "Q4_K_M",
      "format": "GGUF",
      "file_name": "h2o-danube3-500m-chat-Q4_K_M.gguf",
      "file_size_bytes": 352000000,
      "file_size_formatted": "352 MB",
      "sha256": "0f12e8b15d2a9fb9e1987d65b107c390a8fe55102a90184ce1172a8cbf4e1777",
      "download_url": "https://huggingface.co/h2oai/h2o-danube3-500m-chat-GGUF/resolve/main/h2o-danube3-500m-chat-Q4_K_M.gguf",
      "backup_url": "https://cdn.pocketai.pages.dev/models/h2o-danube3-500m-chat-Q4_K_M.gguf",
      "license": "Apache 2.0",
      "license_url": "https://huggingface.co/datasets/choosealicense/licenses/blob/main/markdown/apache-2.0.md",
      "chat_template": "chatml",
      "context_window_default": 2048,
      "context_window_max": 4096,
      "min_ram_gb": 3,
      "recommended_ram_gb": 4,
      "is_pro": false,
      "is_starter": false,
      "capabilities": [
        "Edge Classification",
        "Short Note Taking",
        "Zero Cloud Reliance",
        "Rapid Responses"
      ],
      "hardware_notes": "Created by Kaggle Grandmasters for ultra-fast on-device responsiveness."
    },
    {
      "id": "llama-3.2-3b-instruct-q4km",
      "name": "Llama 3.2 3B Instruct",
      "creator": "Meta AI",
      "tagline": "Pro-tier conversational powerhouse for nuanced writing and multi-step tasks",
      "parameters": "3.2B",
      "quantization": "Q4_K_M",
      "format": "GGUF",
      "file_name": "Llama-3.2-3B-Instruct-Q4_K_M.gguf",
      "file_size_bytes": 2048901120,
      "file_size_formatted": "1.95 GB",
      "sha256": "df719003507d894b9ca0447347a508f7bfa6d64eb0735747cb9b0ce4a331dfec",
      "download_url": "https://huggingface.co/bartowski/Llama-3.2-3B-Instruct-GGUF/resolve/main/Llama-3.2-3B-Instruct-Q4_K_M.gguf",
      "backup_url": "https://cdn.pocketai.pages.dev/models/Llama-3.2-3B-Instruct-Q4_K_M.gguf",
      "license": "Llama 3.2 Community License",
      "license_url": "https://github.com/meta-llama/llama-models/blob/main/models/llama3_2/LICENSE",
      "chat_template": "llama3",
      "context_window_default": 2048,
      "context_window_max": 8192,
      "min_ram_gb": 6,
      "recommended_ram_gb": 8,
      "is_pro": true,
      "is_starter": false,
      "capabilities": [
        "Complex Reasoning",
        "Nuanced Long-form Writing",
        "Function & Tool Calling",
        "Context Memory Retention"
      ],
      "hardware_notes": "Requires 6GB+ physical RAM. Unlocks highest quality on-device intelligence."
    },
    {
      "id": "deepseek-r1-distill-qwen-1.5b-q4km",
      "name": "DeepSeek R1 Distill 1.5B",
      "creator": "DeepSeek AI",
      "tagline": "Pro reasoning model with explicit step-by-step thinking scratchpad",
      "parameters": "1.5B",
      "quantization": "Q4_K_M",
      "format": "GGUF",
      "file_name": "DeepSeek-R1-Distill-Qwen-1.5B-Q4_K_M.gguf",
      "file_size_bytes": 1184600000,
      "file_size_formatted": "1.15 GB",
      "sha256": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
      "download_url": "https://huggingface.co/bartowski/DeepSeek-R1-Distill-Qwen-1.5B-GGUF/resolve/main/DeepSeek-R1-Distill-Qwen-1.5B-Q4_K_M.gguf",
      "backup_url": "https://cdn.pocketai.pages.dev/models/DeepSeek-R1-Distill-Qwen-1.5B-Q4_K_M.gguf",
      "license": "MIT License",
      "license_url": "https://opensource.org/licenses/MIT",
      "chat_template": "chatml",
      "context_window_default": 2048,
      "context_window_max": 4096,
      "min_ram_gb": 4,
      "recommended_ram_gb": 6,
      "is_pro": true,
      "is_starter": false,
      "capabilities": [
        "<think> Scratchpad Reasoning",
        "Complex Logic & Puzzles",
        "Autonomous Tool Calling",
        "Analytical Problem Solving"
      ],
      "hardware_notes": "Outputs its internal chain-of-thought in real-time before answering."
    },
    {
      "id": "qwen2.5-3b-instruct-q4km",
      "name": "Qwen2.5 3B Instruct",
      "creator": "Alibaba Cloud / Qwen",
      "tagline": "Pro multi-turn conversationalist with deep knowledge across 29 languages",
      "parameters": "3.0B",
      "quantization": "Q4_K_M",
      "format": "GGUF",
      "file_name": "qwen2.5-3b-instruct-q4_k_m.gguf",
      "file_size_bytes": 1980000000,
      "file_size_formatted": "1.98 GB",
      "sha256": "1a23e8b15d2a9fb9e1987d65b107c390a8fe55102a90184ce1172a8cbf4e1778",
      "download_url": "https://huggingface.co/Qwen/Qwen2.5-3B-Instruct-GGUF/resolve/main/qwen2.5-3b-instruct-q4_k_m.gguf",
      "backup_url": "https://cdn.pocketai.pages.dev/models/qwen2.5-3b-instruct-q4_k_m.gguf",
      "license": "Qwen Research License",
      "license_url": "https://huggingface.co/Qwen/Qwen2.5-3B-Instruct",
      "chat_template": "chatml",
      "context_window_default": 2048,
      "context_window_max": 8192,
      "min_ram_gb": 6,
      "recommended_ram_gb": 8,
      "is_pro": true,
      "is_starter": false,
      "capabilities": [
        "Advanced Math & Logic",
        "High Context Coherence",
        "Tool Invocation",
        "Multilingual Proficiency"
      ],
      "hardware_notes": "Spectacular benchmark scores exceeding most 7B models from 2023."
    },
    {
      "id": "qwen2.5-coder-3b-instruct-q4km",
      "name": "Qwen2.5 Coder 3B",
      "creator": "Qwen Code Team",
      "tagline": "Full-stack code generation, refactoring, and automated test writing",
      "parameters": "3.1B",
      "quantization": "Q4_K_M",
      "format": "GGUF",
      "file_name": "qwen2.5-coder-3b-instruct-q4_k_m.gguf",
      "file_size_bytes": 2050000000,
      "file_size_formatted": "2.05 GB",
      "sha256": "2b34e8b15d2a9fb9e1987d65b107c390a8fe55102a90184ce1172a8cbf4e1779",
      "download_url": "https://huggingface.co/Qwen/Qwen2.5-Coder-3B-Instruct-GGUF/resolve/main/qwen2.5-coder-3b-instruct-q4_k_m.gguf",
      "backup_url": "https://cdn.pocketai.pages.dev/models/qwen2.5-coder-3b-instruct-q4_k_m.gguf",
      "license": "Qwen Research License",
      "license_url": "https://huggingface.co/Qwen/Qwen2.5-Coder-3B-Instruct",
      "chat_template": "chatml",
      "context_window_default": 2048,
      "context_window_max": 8192,
      "min_ram_gb": 6,
      "recommended_ram_gb": 8,
      "is_pro": true,
      "is_starter": false,
      "capabilities": [
        "Code Completion",
        "Full File Refactoring",
        "Terminal Scripts",
        "Offline Dart & Flutter Code"
      ],
      "hardware_notes": "Top tier coding assistant for portable offline programming."
    },
    {
      "id": "ministral-3b-instruct-q4km",
      "name": "Ministral 3B Instruct",
      "creator": "Mistral AI",
      "tagline": "State of the art edge intelligence designed by Mistral AI",
      "parameters": "3.0B",
      "quantization": "Q4_K_M",
      "format": "GGUF",
      "file_name": "Ministral-3b-instruct-Q4_K_M.gguf",
      "file_size_bytes": 2100000000,
      "file_size_formatted": "2.10 GB",
      "sha256": "3c45e8b15d2a9fb9e1987d65b107c390a8fe55102a90184ce1172a8cbf4e1780",
      "download_url": "https://huggingface.co/bartowski/Ministral-3b-instruct-GGUF/resolve/main/Ministral-3b-instruct-Q4_K_M.gguf",
      "backup_url": "https://cdn.pocketai.pages.dev/models/Ministral-3b-instruct-Q4_K_M.gguf",
      "license": "Mistral Research License",
      "license_url": "https://mistral.ai/news/ministral-3b/",
      "chat_template": "mistral",
      "context_window_default": 2048,
      "context_window_max": 8192,
      "min_ram_gb": 6,
      "recommended_ram_gb": 8,
      "is_pro": true,
      "is_starter": false,
      "capabilities": [
        "Nuanced Dialogue",
        "Creative Prose",
        "Structured Extraction",
        "High General Knowledge"
      ],
      "hardware_notes": "Mistral's flagship on-device architecture with interleaved sliding window."
    },
    {
      "id": "phi-3.5-mini-instruct-q4km",
      "name": "Phi-3.5 Mini Instruct",
      "creator": "Microsoft",
      "tagline": "Trained on synthetic textbooks for exceptionally dense STEM reasoning",
      "parameters": "3.8B",
      "quantization": "Q4_K_M",
      "format": "GGUF",
      "file_name": "Phi-3.5-mini-instruct-Q4_K_M.gguf",
      "file_size_bytes": 2390000000,
      "file_size_formatted": "2.39 GB",
      "sha256": "4d56e8b15d2a9fb9e1987d65b107c390a8fe55102a90184ce1172a8cbf4e1781",
      "download_url": "https://huggingface.co/bartowski/Phi-3.5-mini-instruct-GGUF/resolve/main/Phi-3.5-mini-instruct-Q4_K_M.gguf",
      "backup_url": "https://cdn.pocketai.pages.dev/models/Phi-3.5-mini-instruct-Q4_K_M.gguf",
      "license": "MIT License",
      "license_url": "https://opensource.org/licenses/MIT",
      "chat_template": "phi3",
      "context_window_default": 2048,
      "context_window_max": 8192,
      "min_ram_gb": 6,
      "recommended_ram_gb": 8,
      "is_pro": true,
      "is_starter": false,
      "capabilities": [
        "Math & Physics Solving",
        "Logic Puzzles",
        "Textbook Quality Prose",
        "Data Analysis"
      ],
      "hardware_notes": "Rivals 14B models on STEM benchmarks due to curated synthetic training."
    },
    {
      "id": "phi-4-mini-instruct-q4km",
      "name": "Phi-4 Mini 3.8B",
      "creator": "Microsoft",
      "tagline": "Microsoft's latest generation compact model with state-of-the-art reasoning",
      "parameters": "3.8B",
      "quantization": "Q4_K_M",
      "format": "GGUF",
      "file_name": "Phi-4-mini-instruct-Q4_K_M.gguf",
      "file_size_bytes": 2490000000,
      "file_size_formatted": "2.49 GB",
      "sha256": "5e67e8b15d2a9fb9e1987d65b107c390a8fe55102a90184ce1172a8cbf4e1782",
      "download_url": "https://huggingface.co/bartowski/microsoft_Phi-4-mini-instruct-GGUF/resolve/main/microsoft_Phi-4-mini-instruct-Q4_K_M.gguf",
      "backup_url": "https://cdn.pocketai.pages.dev/models/Phi-4-mini-instruct-Q4_K_M.gguf",
      "license": "MIT License",
      "license_url": "https://opensource.org/licenses/MIT",
      "chat_template": "phi3",
      "context_window_default": 2048,
      "context_window_max": 8192,
      "min_ram_gb": 6,
      "recommended_ram_gb": 8,
      "is_pro": true,
      "is_starter": false,
      "capabilities": [
        "Complex Analytical Reasoning",
        "Coding & Algorithms",
        "High Instruction Adherence",
        "Step-by-Step Proofs"
      ],
      "hardware_notes": "Next-generation Phi architecture featuring enhanced multilingual support."
    },
    {
      "id": "danube3-4b-instruct-q4km",
      "name": "H2O Danube3 4B",
      "creator": "H2O.ai",
      "tagline": "Dense instruction-tuned edge model designed for executive summaries",
      "parameters": "4.0B",
      "quantization": "Q4_K_M",
      "format": "GGUF",
      "file_name": "h2o-danube3-4b-chat-Q4_K_M.gguf",
      "file_size_bytes": 2600000000,
      "file_size_formatted": "2.60 GB",
      "sha256": "6f78e8b15d2a9fb9e1987d65b107c390a8fe55102a90184ce1172a8cbf4e1783",
      "download_url": "https://huggingface.co/h2oai/h2o-danube3-4b-chat-GGUF/resolve/main/h2o-danube3-4b-chat-Q4_K_M.gguf",
      "backup_url": "https://cdn.pocketai.pages.dev/models/h2o-danube3-4b-chat-Q4_K_M.gguf",
      "license": "Apache 2.0",
      "license_url": "https://huggingface.co/datasets/choosealicense/licenses/blob/main/markdown/apache-2.0.md",
      "chat_template": "chatml",
      "context_window_default": 2048,
      "context_window_max": 8192,
      "min_ram_gb": 6,
      "recommended_ram_gb": 8,
      "is_pro": true,
      "is_starter": false,
      "capabilities": [
        "Executive Summaries",
        "Report Writing",
        "Contract & Legal Analysis",
        "High Precision"
      ],
      "hardware_notes": "Apache 2.0 open weights. Great balance of quality and speed."
    },
    {
      "id": "openhermes-2.5-mistral-7b-q3km",
      "name": "OpenHermes 2.5 Mistral 7B",
      "creator": "Teknium",
      "tagline": "Legendary uncensored creative assistant trained on 1M diverse instructions",
      "parameters": "7.2B",
      "quantization": "Q3_K_M",
      "format": "GGUF",
      "file_name": "openhermes-2.5-mistral-7b.Q3_K_M.gguf",
      "file_size_bytes": 3520000000,
      "file_size_formatted": "3.52 GB",
      "sha256": "7a89e8b15d2a9fb9e1987d65b107c390a8fe55102a90184ce1172a8cbf4e1784",
      "download_url": "https://huggingface.co/TheBloke/OpenHermes-2.5-Mistral-7B-GGUF/resolve/main/openhermes-2.5-mistral-7b.Q3_K_M.gguf",
      "backup_url": "https://cdn.pocketai.pages.dev/models/openhermes-2.5-mistral-7b.Q3_K_M.gguf",
      "license": "Apache 2.0",
      "license_url": "https://huggingface.co/datasets/choosealicense/licenses/blob/main/markdown/apache-2.0.md",
      "chat_template": "chatml",
      "context_window_default": 2048,
      "context_window_max": 8192,
      "min_ram_gb": 8,
      "recommended_ram_gb": 8,
      "is_pro": true,
      "is_starter": false,
      "capabilities": [
        "Creative Fiction & Roleplay",
        "Unfiltered Brainstorming",
        "Philosophy & Poetry",
        "Deep Narrative"
      ],
      "hardware_notes": "Requires high-end phone with 8GB+ RAM. 7B parameter richness in Q3_K_M."
    },
    {
      "id": "deepseek-r1-distill-llama-8b-q4km",
      "name": "DeepSeek R1 Distill 8B",
      "creator": "DeepSeek AI",
      "tagline": "Flagship mathematical & coding titan with full reasoning chain",
      "parameters": "8.0B",
      "quantization": "Q4_K_M",
      "format": "GGUF",
      "file_name": "DeepSeek-R1-Distill-Llama-8B-Q4_K_M.gguf",
      "file_size_bytes": 4920000000,
      "file_size_formatted": "4.92 GB",
      "sha256": "8b90e8b15d2a9fb9e1987d65b107c390a8fe55102a90184ce1172a8cbf4e1785",
      "download_url": "https://huggingface.co/bartowski/DeepSeek-R1-Distill-Llama-8B-GGUF/resolve/main/DeepSeek-R1-Distill-Llama-8B-Q4_K_M.gguf",
      "backup_url": "https://cdn.pocketai.pages.dev/models/DeepSeek-R1-Distill-Llama-8B-Q4_K_M.gguf",
      "license": "MIT License",
      "license_url": "https://opensource.org/licenses/MIT",
      "chat_template": "llama3",
      "context_window_default": 2048,
      "context_window_max": 8192,
      "min_ram_gb": 8,
      "recommended_ram_gb": 12,
      "is_pro": true,
      "is_starter": false,
      "capabilities": [
        "Unrivalled Offline Math & Science",
        "Autonomous Deep Reasoning",
        "Competitive Programming",
        "Long Chain-of-Thought"
      ],
      "hardware_notes": "For flagship devices (Snapdragon 8 Gen 2/3, 8GB-12GB RAM). Unmatched intelligence."
    }
  ]
}