{
  "checked": "2026-09-13",
  "guide": "https://share.directivecreator.com/wemustpacethefrontier/",
  "models": [
    {
      "id": "glm",
      "name": "GLM-5.3-Flash",
      "maker": "Z.ai",
      "tag": "Highest reference capability",
      "score": 42,
      "size": 199.71,
      "memory": 256,
      "hardware": "256 GB unified-memory workstation; a tight fit",
      "params": "320B total · 18B active",
      "context": "1,048,576 tokens",
      "license": "MIT",
      "description": "The strongest reference benchmark result in this selection. A high-end choice for difficult coding, reasoning and multi-step work.",
      "tradeoff": "A roughly 200 GB download and emerging runtime support make this an enthusiast project, not a normal laptop install.",
      "repo": "zai-org/GLM-5.3-Flash",
      "package": "https://huggingface.co/unsloth/GLM-5.3-Flash-GGUF/tree/main/UD-Q4_K_XL",
      "quant": "UD-Q4_K_XL · community GGUF",
      "runtime": "Unsloth Desktop",
      "setup": "Install Unsloth Desktop, search unsloth/GLM-5.3-Flash-GGUF, select UD-Q4_K_XL, then download and load with 8K context.",
      "command": "",
      "guide": "https://unsloth.ai/docs/models/glm-5.3-flash",
      "extra": "Text-chat starting route. The optional vision projector adds about 1.16 GB. Follow the model guide for its required build; generic older llama.cpp releases may not load this architecture.",
      "estimated": false,
      "benchmark": "https://artificialanalysis.ai/models/glm-5-3-flash",
      "official": "https://huggingface.co/zai-org/GLM-5.3-Flash",
      "license_url": "https://huggingface.co/zai-org/GLM-5.3-Flash"
    },
    {
      "id": "flash-next",
      "name": "Qwen3.8-Flash-Next",
      "maker": "Qwen",
      "tag": "Advanced Apple Silicon",
      "score": 40,
      "size": 104.85,
      "memory": 192,
      "hardware": "192 GB+ Apple Silicon; 128 GB is tight",
      "params": "125B backbone · 6B active · ~180B full package",
      "context": "262,144 native · up to 1M extended",
      "license": "Qwen Community 1.0",
      "description": "Strong coding and tool-use capability with a smaller download than the biggest workstation models.",
      "tradeoff": "Experimental architecture and restricted commercial terms. The selected MLX package requires Apple Silicon.",
      "repo": "Qwen/Qwen3.8-Flash-Next",
      "package": "https://ollama.com/library/qwen3.8-flash-next:125b-mlx",
      "quant": "125b-mlx · Ollama package",
      "runtime": "Ollama",
      "setup": "Install current Ollama on Apple Silicon. Set context to 8K in settings before running the command.",
      "command": "ollama run qwen3.8-flash-next:125b-mlx",
      "guide": "https://ollama.com/library/qwen3.8-flash-next:125b-mlx",
      "extra": "The full package includes ~51B n-gram embeddings and ~4B prediction parameters beyond its 125B backbone. The publisher license restricts commercial model-service and AI work-assistant businesses; an internal-use exception and large-scale attribution conditions apply. Read the publisher license, even if a package label says Apache.",
      "estimated": false,
      "benchmark": "https://artificialanalysis.ai/models/qwen3-8-flash-next",
      "official": "https://huggingface.co/Qwen/Qwen3.8-Flash-Next",
      "license_url": "https://huggingface.co/Qwen/Qwen3.8-Flash-Next/blob/main/LICENSE"
    },
    {
      "id": "deepseek",
      "name": "DeepSeek-V4-Flash-0731",
      "maker": "DeepSeek",
      "tag": "Scientific reasoning alternative",
      "score": 35,
      "size": 161.87,
      "memory": 256,
      "hardware": "256 GB unified-memory workstation",
      "params": "284B total · 13B active",
      "context": "1,048,576 tokens",
      "license": "MIT",
      "description": "A strong reasoning and coding alternative, especially worth evaluating on scientific problems. The dated 0731 checkpoint is the one compared here.",
      "tradeoff": "Long thinking sequences can be slow locally. Its 13B active count does not make the whole model small.",
      "repo": "deepseek-ai/DeepSeek-V4-Flash-0731",
      "package": "https://huggingface.co/unsloth/DeepSeek-V4-Flash-0731-GGUF/tree/main/UD-Q8_K_XL",
      "quant": "UD-Q8_K_XL · community GGUF",
      "runtime": "llama.cpp",
      "setup": "Install a current llama.cpp build. Run this text-chat server, then open http://127.0.0.1:8080 after loading.",
      "command": "llama-server -hf unsloth/DeepSeek-V4-Flash-0731-GGUF:UD-Q8_K_XL -c 8192 --jinja --temp 1.0 --top-p 1.0 --host 127.0.0.1 --port 8080",
      "guide": "https://unsloth.ai/docs/models/deepseek-v4",
      "extra": "Mixed-precision source weights mean the Q8 name does not imply eight bits for every parameter. Start without the optional draft model. The reference score uses max reasoning effort; this starter command does not reproduce that evaluation.",
      "estimated": false,
      "benchmark": "https://artificialanalysis.ai/models/deepseek-v4-flash",
      "official": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731",
      "license_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731"
    },
    {
      "id": "qwen27",
      "name": "Qwen3.8-27B",
      "maker": "Qwen",
      "tag": "Best everyday starting point",
      "score": 34,
      "size": 17.74,
      "memory": 32,
      "hardware": "32–48 GB unified memory, or 24–32 GB VRAM",
      "params": "27B dense language model + vision",
      "context": "262,144 native · up to 1M extended",
      "license": "Apache-2.0",
      "description": "The practical first pick for writing, coding, reasoning and image understanding. A strong balance of capability, size and straightforward installation.",
      "tradeoff": "Less capable than the big leaders on demanding agentic work. Thinking can be verbose, and a 24 GB GPU gets tight as context grows.",
      "repo": "Qwen/Qwen3.8-27B",
      "package": "https://ollama.com/library/qwen3.8:27b-q4_K_M",
      "quant": "27b-q4_K_M · Ollama package",
      "runtime": "Ollama",
      "setup": "Install current Ollama. Set context to 8K in settings, then paste this into your terminal. The first run downloads the model.",
      "command": "ollama run qwen3.8:27b-q4_K_M",
      "guide": "https://ollama.com/library/qwen3.8:27b-q4_K_M",
      "extra": "Allow roughly 20–26 GB for the model process at modest text context, plus memory for your OS and apps. This is an estimate, not a measured peak. The download includes the packaged vision component. Reference score uses xhigh effort.",
      "estimated": false,
      "benchmark": "https://artificialanalysis.ai/models/qwen3-8-27b",
      "official": "https://huggingface.co/Qwen/Qwen3.8-27B",
      "license_url": "https://huggingface.co/Qwen/Qwen3.8-27B"
    },
    {
      "id": "inkling",
      "name": "Inkling-Small",
      "maker": "Thinking Machines",
      "tag": "Multimodal workstation option",
      "score": 26,
      "size": 163.27,
      "memory": 256,
      "hardware": "256 GB unified-memory workstation",
      "params": "276B total · 12B active (publisher figures)",
      "context": "1M tokens in native configuration",
      "license": "Apache-2.0",
      "description": "A generalist with text, image and audio inputs in its native stack. Useful to explore if you already have a large workstation.",
      "tradeoff": "“Small” is a family name, not a laptop footprint. Local applications may expose text only, and runtime support is still developing.",
      "repo": "thinkingmachines/Inkling-Small",
      "package": "https://huggingface.co/unsloth/Inkling-Small-GGUF/tree/main/UD-Q4_K_XL",
      "quant": "UD-Q4_K_XL · community GGUF",
      "runtime": "Unsloth Desktop",
      "setup": "Install current Unsloth Desktop. Search unsloth/Inkling-Small-GGUF, select UD-Q4_K_XL and load with 8K context. Start with text.",
      "command": "",
      "guide": "https://unsloth.ai/docs/models/inkling",
      "extra": "Use the model-specific supported app/build. The publisher reports 276B while the repository tensor counter reports about 266B; actual file totals determine this download size. Native modalities do not guarantee local wrapper support.",
      "estimated": false,
      "benchmark": "https://artificialanalysis.ai/models/inkling-small",
      "official": "https://huggingface.co/thinkingmachines/Inkling-Small",
      "license_url": "https://huggingface.co/thinkingmachines/Inkling-Small"
    },
    {
      "id": "ling",
      "name": "Ling-3.0-flash",
      "maker": "inclusionAI",
      "tag": "Middle workstation tier",
      "score": 25,
      "size": 77.01,
      "memory": 128,
      "hardware": "128 GB unified-memory workstation",
      "params": "124B total · 5.1B active",
      "context": "262,144 tokens",
      "license": "MIT",
      "description": "An efficient mixture-of-experts architecture aimed at coding, research and tool workflows, with a smaller footprint than most workstation choices.",
      "tradeoff": "The benchmark score is estimated. Promising placement, but weaker evidence than a completed independent evaluation.",
      "repo": "inclusionAI/Ling-3.0-flash",
      "package": "https://huggingface.co/inclusionAI/Ling-3.0-flash-GGUF/tree/main/Q4_K_M",
      "quant": "Q4_K_M · publisher GGUF",
      "runtime": "llama.cpp",
      "setup": "Install current llama.cpp with Ling architecture support. Run the server, then open http://127.0.0.1:8080 after loading.",
      "command": "llama-server -hf inclusionAI/Ling-3.0-flash-GGUF:Q4_K_M -c 8192 --jinja --temp 0.6 --top-p 0.95 --host 127.0.0.1 --port 8080",
      "guide": "https://huggingface.co/inclusionAI/Ling-3.0-flash-GGUF",
      "extra": "Download size includes both model shards. This is the text model. Start around 8K context; the full advertised context needs substantially more working memory.",
      "estimated": true,
      "benchmark": "https://artificialanalysis.ai/models/ling-3-0-flash",
      "official": "https://huggingface.co/inclusionAI/Ling-3.0-flash",
      "license_url": "https://huggingface.co/inclusionAI/Ling-3.0-flash"
    },
    {
      "id": "minimax",
      "name": "MiniMax-M2.7",
      "maker": "MiniMax",
      "tag": "Personal coding experiments",
      "score": 23,
      "size": 140.78,
      "memory": 192,
      "hardware": "192–256 GB memory; 256 GB offers more room",
      "params": "230B total · 10B active",
      "context": "~200K · config 204,800",
      "license": "Custom · commercial permission required",
      "description": "A coding and tool-use model for personal self-hosted work. An option for people who want to compare a different model family.",
      "tradeoff": "Substantial memory needs. Commercial use requires prior written authorization; this is not ordinary MIT licensing.",
      "repo": "MiniMaxAI/MiniMax-M2.7",
      "package": "https://huggingface.co/unsloth/MiniMax-M2.7-GGUF/tree/main/UD-Q4_K_XL",
      "quant": "UD-Q4_K_XL · community GGUF",
      "runtime": "llama.cpp",
      "setup": "For personal use, install current llama.cpp and run this text-chat server. Open http://127.0.0.1:8080 after loading.",
      "command": "llama-server -hf unsloth/MiniMax-M2.7-GGUF:UD-Q4_K_XL -c 8192 --jinja --temp 1.0 --top-p 0.95 --top-k 40 --host 127.0.0.1 --port 8080",
      "guide": "https://unsloth.ai/docs/models/tutorials/minimax-m27",
      "extra": "The local guide lists 196,608 context versus 204,800 in config. Start at 8K. Read the full publisher license for attribution and use conditions before adopting it for work.",
      "estimated": false,
      "benchmark": "https://artificialanalysis.ai/models/minimax-m2-7",
      "official": "https://huggingface.co/MiniMaxAI/MiniMax-M2.7",
      "license_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.7/blob/main/LICENSE"
    },
    {
      "id": "mimo",
      "name": "MiMo-V2.5",
      "maker": "Xiaomi",
      "tag": "Large multimodal generalist",
      "score": 22,
      "size": 191.61,
      "memory": 256,
      "hardware": "256 GB unified-memory workstation; limited headroom",
      "params": "310B total · 15B active",
      "context": "1,048,576 tokens",
      "license": "MIT",
      "description": "Built for long-context and agent workflows, with text, image, video and audio support in its full native stack.",
      "tradeoff": "A very large download for its reference score. The simple local command below is text chat, not a promise of full audio/video support.",
      "repo": "XiaomiMiMo/MiMo-V2.5",
      "package": "https://huggingface.co/unsloth/MiMo-V2.5-GGUF/tree/main/UD-Q4_K_XL",
      "quant": "UD-Q4_K_XL · community GGUF",
      "runtime": "llama.cpp",
      "setup": "Install current llama.cpp. Run this server, then open http://127.0.0.1:8080 after loading. Keep --jinja for the chat-template fixes.",
      "command": "llama-server -hf unsloth/MiMo-V2.5-GGUF:UD-Q4_K_XL -c 8192 --jinja --temp 1.0 --top-p 0.95 --host 127.0.0.1 --port 8080",
      "guide": "https://huggingface.co/unsloth/MiMo-V2.5-GGUF",
      "extra": "The selected conversion does not bundle a separate vision projector. Use updated runtime and quantization files; native model features and features in this local setup differ.",
      "estimated": false,
      "benchmark": "https://artificialanalysis.ai/models/mimo-v2-5-0424",
      "official": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5",
      "license_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5"
    },
    {
      "id": "qwen9",
      "name": "Qwen3.5-9B",
      "maker": "Qwen",
      "tag": "Best starting point for 16 GB",
      "score": 14,
      "size": 6.59,
      "memory": 16,
      "hardware": "16 GB unified memory, or ~12 GB VRAM + adequate RAM",
      "params": "9B dense model",
      "context": "262,144 tokens",
      "license": "Apache-2.0",
      "description": "The smallest download here. A sensible way to begin local writing, chat and lighter coding without buying a workstation.",
      "tradeoff": "Lower reference capability than the larger models. Complex reasoning and long conversations are harder; its score is estimated.",
      "repo": "Qwen/Qwen3.5-9B",
      "package": "https://ollama.com/library/qwen3.5:9b",
      "quant": "9b · Ollama Q4_K_M package",
      "runtime": "Ollama",
      "setup": "Install Ollama. Begin with 4K–8K context in settings and run the command below. Keep other memory-heavy apps closed.",
      "command": "ollama run qwen3.5:9b",
      "guide": "https://ollama.com/library/qwen3.5:9b",
      "extra": "Plan for roughly 9–12 GB for the model process at modest context, plus your OS. That is an estimate. CPU-only execution can work with enough RAM, but expect slower answers.",
      "estimated": true,
      "benchmark": "https://artificialanalysis.ai/models/qwen3-5-9b",
      "official": "https://huggingface.co/Qwen/Qwen3.5-9B",
      "license_url": "https://huggingface.co/Qwen/Qwen3.5-9B"
    },
    {
      "id": "orca",
      "name": "Qwen3.8-27B-Uncensored",
      "maker": "OrcaRouter",
      "tag": "Requested specialist inclusion",
      "score": null,
      "size": 17.74,
      "memory": 32,
      "hardware": "32–48 GB unified memory, or 24–32 GB VRAM",
      "params": "27B dense derivative + vision",
      "context": "262,144 advertised for the model family",
      "license": "Apache-2.0 + publisher use conditions",
      "description": "A weight-edited Qwen derivative intended to reduce refusal behavior. Included for controlled research and comparison, not as a capability upgrade.",
      "tradeoff": "No comparable independent score verified for this exact derivative. Reduced refusals do not mean better factual accuracy; read its research-use conditions.",
      "repo": "orcarouter/Qwen3.8-27B-Uncensored-GGUF",
      "package": "https://ollama.com/orcarouter/Qwen3.8-27B-Uncensored:q4_K_M",
      "quant": "q4_K_M · OrcaRouter Ollama package",
      "runtime": "Ollama",
      "setup": "Read OrcaRouter’s model conditions first. For its local quantized package, install Ollama, set context to 8K, then run:",
      "command": "ollama run orcarouter/Qwen3.8-27B-Uncensored:q4_K_M",
      "guide": "https://www.orcarouter.ai/blog/qwen-uncensored-explained",
      "extra": "Selected interpretation of “OrcaRouter Qwen uncensored”: the publisher’s guide points to this 27B family. Hugging Face downloads may require sign-in and acceptance. The hosted obsidian/Qwen3.8-27B service is separate from this Q4 local package; hosted availability and access conditions can change.",
      "estimated": false,
      "benchmark": null,
      "official": "https://huggingface.co/orcarouter/Qwen3.8-27B-Uncensored-GGUF",
      "license_url": "https://huggingface.co/orcarouter/Qwen3.8-27B-Uncensored-GGUF"
    }
  ]
}