Variable MODELSConst

Type Declaration

• Readonlygemma-3n-e2b: {
      name: "Gemma 3n E2B";
      repo: "unsloth/gemma-3n-E2B-it-GGUF";
      file: "gemma-3n-E2B-it-Q4_K_M.gguf";
      parameters: "5B→2B";
      quantization: "Q4_K_M";
      contextLength: 32768;
      languages: readonly [
          "en",
          "de",
          "fr",
          "es",
          "it",
          "pt",
          "nl",
          "pl",
          "ru",
          "ja",
          "ko",
          "zh",
      ];
      description: "Ultra-efficient edge model, ~2GB RAM";
      requiresAuth: false;
      benchmarks: { mmlu: 64; arena: 1250 };
  }

• Readonlygemma-3n-e4b: {
      name: "Gemma 3n E4B";
      repo: "unsloth/gemma-3n-E4B-it-GGUF";
      file: "gemma-3n-E4B-it-Q4_K_M.gguf";
      parameters: "8B→4B";
      quantization: "Q4_K_M";
      contextLength: 32768;
      languages: readonly [
          "en",
          "de",
          "fr",
          "es",
          "it",
          "pt",
          "nl",
          "pl",
          "ru",
          "ja",
          "ko",
          "zh",
      ];
      description: "Best edge model, ~3GB RAM";
      requiresAuth: false;
      benchmarks: { mmlu: 75; arena: 1300 };
  }

• Readonlygemma-3-27b: {
      name: "Gemma 3 27B";
      repo: "unsloth/gemma-3-27b-it-GGUF";
      file: "gemma-3-27b-it-Q4_K_M.gguf";
      parameters: "27B";
      quantization: "Q4_K_M";
      contextLength: 131072;
      languages: readonly [
          "en",
          "de",
          "fr",
          "es",
          "it",
          "pt",
          "nl",
          "pl",
          "ru",
          "ja",
          "ko",
          "zh",
      ];
      description: "Maximum quality, 128K context, ~18GB RAM";
      benchmarks: { mmlu: 77; arena: 1338 };
  }

• Readonlygpt-oss-20b: {
      name: "GPT-OSS 20B";
      repo: "unsloth/gpt-oss-20b-GGUF";
      file: "gpt-oss-20b-Q4_K_M.gguf";
      parameters: "21B (3.6B active)";
      quantization: "Q4_K_M";
      contextLength: 131072;
      languages: readonly ["en"];
      description: "OpenAI's open model, MoE, ~16GB RAM";
      benchmarks: { mmlu: 82; arena: 1340 };
  }

• Readonlyphi-4: {
      name: "Phi-4 14B";
      repo: "bartowski/phi-4-GGUF";
      file: "phi-4-Q4_K_M.gguf";
      parameters: "14B";
      quantization: "Q4_K_M";
      contextLength: 16384;
      languages: readonly ["en"];
      description: "Microsoft's reasoning-focused, excellent for STEM";
      benchmarks: { mmlu: 84; arena: 1320 };
  }

• Readonlyqwen3-4b: {
      name: "Qwen3 4B";
      repo: "unsloth/Qwen3-4B-GGUF";
      file: "Qwen3-4B-Q4_K_M.gguf";
      parameters: "4B";
      quantization: "Q4_K_M";
      contextLength: 32768;
      languages: readonly [
          "en",
          "zh",
          "de",
          "fr",
          "es",
          "pt",
          "it",
          "nl",
          "pl",
          "ru",
          "ja",
          "ko",
      ];
      description: "Thinking mode, 100+ languages, ~3GB RAM";
      thinkingMode: "qwen";
      benchmarks: { mmlu: 76; arena: 1300 };
  }

• Readonlyqwen3-8b: {
      name: "Qwen3 8B";
      repo: "unsloth/Qwen3-8B-GGUF";
      file: "Qwen3-8B-Q4_K_M.gguf";
      parameters: "8B";
      quantization: "Q4_K_M";
      contextLength: 32768;
      languages: readonly [
          "en",
          "zh",
          "de",
          "fr",
          "es",
          "pt",
          "it",
          "nl",
          "pl",
          "ru",
          "ja",
          "ko",
      ];
      description: "Thinking mode, excellent multilingual, ~5GB RAM";
      thinkingMode: "qwen";
      benchmarks: { mmlu: 81; arena: 1350 };
  }

• Readonlyqwen3-14b: {
      name: "Qwen3 14B";
      repo: "unsloth/Qwen3-14B-GGUF";
      file: "Qwen3-14B-Q4_K_M.gguf";
      parameters: "14B";
      quantization: "Q4_K_M";
      contextLength: 32768;
      languages: readonly [
          "en",
          "zh",
          "de",
          "fr",
          "es",
          "pt",
          "it",
          "nl",
          "pl",
          "ru",
          "ja",
          "ko",
      ];
      description: "Thinking mode, top multilingual, ~9GB RAM";
      thinkingMode: "qwen";
      benchmarks: { mmlu: 84; arena: 1380 };
  }

• Readonlyqwen-2.5-coder-7b: {
      name: "Qwen 2.5 Coder 7B";
      repo: "bartowski/Qwen2.5-Coder-7B-Instruct-GGUF";
      file: "Qwen2.5-Coder-7B-Instruct-Q4_K_M.gguf";
      parameters: "7B";
      quantization: "Q4_K_M";
      contextLength: 131072;
      languages: readonly ["en"];
      description: "Optimized for code generation";
      benchmarks: { mmlu: 66; arena: 1250 };
  }

• Readonlydeepseek-r1-7b: {
      name: "DeepSeek R1 Distill 7B";
      repo: "bartowski/DeepSeek-R1-Distill-Qwen-7B-GGUF";
      file: "DeepSeek-R1-Distill-Qwen-7B-Q4_K_M.gguf";
      parameters: "7B";
      quantization: "Q4_K_M";
      contextLength: 131072;
      languages: readonly ["en", "zh"];
      description: "Strong reasoning with chain-of-thought";
      thinkingMode: "deepseek";
      benchmarks: { mmlu: 72; arena: 1300 };
  }

• Readonlydeepseek-r1-14b: {
      name: "DeepSeek R1 Distill 14B";
      repo: "bartowski/DeepSeek-R1-Distill-Qwen-14B-GGUF";
      file: "DeepSeek-R1-Distill-Qwen-14B-Q4_K_M.gguf";
      parameters: "14B";
      quantization: "Q4_K_M";
      contextLength: 131072;
      languages: readonly ["en", "zh"];
      description: "Best reasoning model, shows thinking";
      thinkingMode: "deepseek";
      benchmarks: { mmlu: 79; arena: 1350 };
  }