{
  "_generated": "data/models.json via scripts/build-models.js — do not edit",
  "last_updated": "2026-11-18",
  "last_checked": "2026-11-18",
  "models": [
    {
      "id": "llama-4-maverick",
      "name": "Llama 4 Maverick",
      "provider": "Meta",
      "params": 400,
      "active_params": 17,
      "is_moe": true,
      "kv_head_fraction": 0.2,
      "arch": {
        "layers": 48,
        "hidden": 5120,
        "intermediate": 8192,
        "heads": 40,
        "kv_heads": 8,
        "head_dim": 128,
        "vocab": 202048,
        "tie_embeddings": false,
        "max_position": 1048576,
        "experts": 128,
        "experts_per_tok": 1,
        "sliding_window": 8192,
        "sliding_layers": 36
      },
      "ollama_tag": "llama4:maverick",
      "hf_repo": "meta-llama/Llama-4-Maverick-17B-128E-Instruct",
      "modalities": [
        "text",
        "image"
      ],
      "note": "MoE — 17B active of 400B total, 128 experts. Multimodal (text+image)",
      "release_date": "2025-04",
      "popularity": 7
    },
    {
      "id": "gpt-oss-120b",
      "name": "GPT-oss 120B",
      "provider": "OpenAI",
      "params": 117,
      "active_params": 5.1,
      "is_moe": true,
      "kv_head_fraction": 0.125,
      "arch": {
        "layers": 36,
        "hidden": 2880,
        "intermediate": 2880,
        "heads": 64,
        "kv_heads": 8,
        "head_dim": 64,
        "vocab": 201088,
        "tie_embeddings": false,
        "max_position": 131072,
        "experts": 128,
        "experts_per_tok": 4,
        "sliding_window": 128,
        "sliding_layers": 18
      },
      "ollama_tag": null,
      "hf_repo": "openai/gpt-oss-120b",
      "modalities": [
        "text"
      ],
      "note": null,
      "release_date": "2025-08",
      "popularity": 77
    },
    {
      "id": "glm-5",
      "name": "GLM-5",
      "provider": "Zhipu AI",
      "params": 754,
      "active_params": 40,
      "is_moe": true,
      "kv_head_fraction": null,
      "arch": {
        "layers": 78,
        "hidden": 6144,
        "intermediate": 12288,
        "heads": 64,
        "kv_heads": 64,
        "head_dim": 64,
        "vocab": 154880,
        "tie_embeddings": false,
        "max_position": 202752,
        "kv_lora_rank": 512,
        "qk_rope_head_dim": 64,
        "experts": 256,
        "experts_per_tok": 8,
        "moe_intermediate": 2048,
        "shared_experts": 1
      },
      "ollama_tag": null,
      "hf_repo": "zai-org/GLM-5",
      "modalities": [
        "text"
      ],
      "note": null,
      "release_date": "2026-02",
      "popularity": 65
    },
    {
      "id": "kimi-k2-5",
      "name": "Kimi K2.5",
      "provider": "Moonshot AI",
      "params": 1000,
      "active_params": 32,
      "is_moe": true,
      "kv_head_fraction": null,
      "arch": {
        "layers": 61,
        "hidden": 7168,
        "intermediate": 18432,
        "heads": 64,
        "kv_heads": 64,
        "vocab": 163840,
        "tie_embeddings": false,
        "max_position": 262144,
        "kv_lora_rank": 512,
        "qk_rope_head_dim": 64,
        "experts": 384,
        "experts_per_tok": 8,
        "moe_intermediate": 2048,
        "shared_experts": 1
      },
      "ollama_tag": null,
      "hf_repo": "moonshotai/Kimi-K2.5",
      "modalities": [
        "text"
      ],
      "note": null,
      "release_date": "2026-01",
      "popularity": 50
    },
    {
      "id": "minimax-m2-5",
      "name": "MiniMax M2.5",
      "provider": "MiniMax",
      "params": 230,
      "active_params": null,
      "is_moe": false,
      "kv_head_fraction": 0.1667,
      "arch": {
        "layers": 62,
        "hidden": 3072,
        "intermediate": 1536,
        "heads": 48,
        "kv_heads": 8,
        "head_dim": 128,
        "vocab": 200064,
        "tie_embeddings": false,
        "max_position": 196608,
        "experts": 256,
        "experts_per_tok": 8
      },
      "ollama_tag": null,
      "hf_repo": "MiniMaxAI/MiniMax-M2.5",
      "modalities": [
        "text"
      ],
      "note": null,
      "release_date": "2026-02",
      "popularity": 62
    },
    {
      "id": "step-3-5-flash",
      "name": "Step-3.5-Flash",
      "provider": "Stepfun",
      "params": 196,
      "active_params": 11,
      "is_moe": true,
      "kv_head_fraction": 0.125,
      "arch": {
        "layers": 45,
        "hidden": 4096,
        "intermediate": 11264,
        "heads": 64,
        "kv_heads": 8,
        "head_dim": 128,
        "vocab": 128896,
        "tie_embeddings": false,
        "max_position": 262144,
        "experts": 288,
        "moe_intermediate": 1280,
        "sliding_window": 512,
        "sliding_layers": 36
      },
      "ollama_tag": null,
      "hf_repo": "stepfun-ai/Step-3.5-Flash",
      "modalities": [
        "text"
      ],
      "note": null,
      "release_date": "2026-02",
      "popularity": 31
    },
    {
      "id": "llama-4-scout",
      "name": "Llama 4 Scout",
      "provider": "Meta",
      "params": 109,
      "active_params": 17,
      "is_moe": true,
      "kv_head_fraction": 0.2,
      "arch": {
        "layers": 48,
        "hidden": 5120,
        "intermediate": 8192,
        "heads": 40,
        "kv_heads": 8,
        "head_dim": 128,
        "vocab": 202048,
        "tie_embeddings": false,
        "max_position": 10485760,
        "experts": 16,
        "experts_per_tok": 1,
        "sliding_window": 8192,
        "sliding_layers": 36
      },
      "ollama_tag": "llama4:scout",
      "hf_repo": "meta-llama/Llama-4-Scout-17B-16E-Instruct",
      "modalities": [
        "text",
        "image"
      ],
      "note": "MoE — 17B active of 109B total, 16 experts. Multimodal (text+image)",
      "release_date": "2025-04",
      "popularity": 23
    },
    {
      "id": "deepseek-v3",
      "name": "DeepSeek V3",
      "provider": "DeepSeek",
      "params": 671,
      "active_params": 37,
      "is_moe": true,
      "kv_head_fraction": null,
      "arch": {
        "layers": 61,
        "hidden": 7168,
        "intermediate": 18432,
        "heads": 128,
        "kv_heads": 128,
        "vocab": 129280,
        "tie_embeddings": false,
        "max_position": 163840,
        "kv_lora_rank": 512,
        "qk_rope_head_dim": 64,
        "experts": 256,
        "experts_per_tok": 8,
        "moe_intermediate": 2048,
        "shared_experts": 1
      },
      "ollama_tag": "deepseek-v3",
      "hf_repo": "deepseek-ai/DeepSeek-V3",
      "modalities": [
        "text"
      ],
      "note": null,
      "release_date": "2024-12",
      "popularity": 68
    },
    {
      "id": "deepseek-v3-2",
      "name": "DeepSeek V3.2",
      "provider": "DeepSeek",
      "params": 685,
      "active_params": 37,
      "is_moe": true,
      "kv_head_fraction": null,
      "arch": {
        "layers": 61,
        "hidden": 7168,
        "intermediate": 18432,
        "heads": 128,
        "kv_heads": 128,
        "vocab": 129280,
        "tie_embeddings": false,
        "max_position": 163840,
        "kv_lora_rank": 512,
        "qk_rope_head_dim": 64,
        "experts": 256,
        "experts_per_tok": 8,
        "moe_intermediate": 2048,
        "shared_experts": 1
      },
      "ollama_tag": "deepseek-v3:latest",
      "hf_repo": "deepseek-ai/DeepSeek-V3.2",
      "modalities": [
        "text"
      ],
      "note": "DeepSeek-V3.2 (Dec 2025). LiveCodeBench 83.3 per official paper (74.1 was the V3.2-Exp variant). First-party API now maps deepseek-chat to V4-Flash; V3.2 remains on third-party providers.",
      "release_date": "2025-12",
      "popularity": 98
    },
    {
      "id": "qwen-3-5",
      "name": "Qwen 3.5",
      "provider": "Alibaba",
      "params": 397,
      "active_params": 17,
      "is_moe": true,
      "kv_head_fraction": 0.0625,
      "arch": {
        "layers": 60,
        "hidden": 4096,
        "heads": 32,
        "kv_heads": 2,
        "head_dim": 256,
        "vocab": 248320,
        "tie_embeddings": false,
        "max_position": 262144,
        "experts": 512,
        "experts_per_tok": 10,
        "moe_intermediate": 1024,
        "shared_experts": 1,
        "linear_layers": 45
      },
      "ollama_tag": "qwen3:235b-a22b",
      "hf_repo": "Qwen/Qwen3.5-397B-A17B",
      "modalities": [
        "text",
        "image"
      ],
      "note": null,
      "release_date": "2026-02",
      "popularity": 56
    },
    {
      "id": "mistral-large",
      "name": "Mistral Large",
      "provider": "Mistral AI",
      "params": 675,
      "active_params": 41,
      "is_moe": true,
      "kv_head_fraction": null,
      "arch": {
        "layers": 61,
        "hidden": 7168,
        "intermediate": 16384,
        "heads": 128,
        "kv_heads": 128,
        "head_dim": 192,
        "vocab": 131072,
        "tie_embeddings": false,
        "max_position": 294912,
        "kv_lora_rank": 512,
        "qk_rope_head_dim": 64,
        "experts": 128,
        "experts_per_tok": 4,
        "moe_intermediate": 4096,
        "shared_experts": 1
      },
      "ollama_tag": "mistral-large",
      "hf_repo": "mistralai/Mistral-Large-3-675B-Instruct-2512",
      "modalities": [
        "text"
      ],
      "note": "Mistral Large 3 (2512) — 675B total / 41B active MoE, Apache 2.0. GPQA Diamond ~44 per independent evals (Mistral card claims 67).",
      "release_date": "2025-12",
      "popularity": 2
    },
    {
      "id": "mimo-v2-flash",
      "name": "MiMo-V2-Flash",
      "provider": "Xiaomi",
      "params": 309,
      "active_params": 15,
      "is_moe": true,
      "kv_head_fraction": 0.0625,
      "arch": {
        "layers": 48,
        "hidden": 4096,
        "intermediate": 16384,
        "heads": 64,
        "kv_heads": 4,
        "head_dim": 192,
        "vocab": 152576,
        "tie_embeddings": false,
        "max_position": 262144,
        "experts": 256,
        "experts_per_tok": 8,
        "moe_intermediate": 2048,
        "sliding_window": 128,
        "sliding_layers": 39
      },
      "ollama_tag": null,
      "hf_repo": "XiaomiMiMo/MiMo-V2-Flash",
      "modalities": [
        "text"
      ],
      "note": null,
      "release_date": "2025-12",
      "popularity": 8
    },
    {
      "id": "nemotron-ultra-253b",
      "name": "Nemotron Ultra 253B",
      "provider": "NVIDIA",
      "params": 253,
      "active_params": null,
      "is_moe": false,
      "kv_head_fraction": 0.0625,
      "arch": {
        "layers": 162,
        "hidden": 16384,
        "heads": 128,
        "kv_heads": 8,
        "vocab": 128256,
        "tie_embeddings": false,
        "max_position": 131072,
        "linear_layers": 98
      },
      "ollama_tag": null,
      "hf_repo": "nvidia/Llama-3_1-Nemotron-Ultra-253B-v1",
      "modalities": [
        "text"
      ],
      "note": null,
      "release_date": "2025-04",
      "popularity": 3
    },
    {
      "id": "gemma-4-31b",
      "name": "Gemma 4 31B",
      "provider": "Google",
      "params": 30.7,
      "active_params": null,
      "is_moe": false,
      "kv_head_fraction": 0.5,
      "arch": {
        "layers": 60,
        "hidden": 5376,
        "intermediate": 21504,
        "heads": 32,
        "kv_heads": 16,
        "head_dim": 256,
        "vocab": 262144,
        "tie_embeddings": true,
        "max_position": 262144,
        "sliding_window": 1024,
        "sliding_layers": 50
      },
      "ollama_tag": "gemma4:31b",
      "hf_repo": "google/gemma-4-31B-it",
      "modalities": [
        "text",
        "image"
      ],
      "note": "Dense. Multimodal, 256K ctx",
      "release_date": "2026-04",
      "popularity": 94
    },
    {
      "id": "llama-3-2-1b",
      "name": "Llama 3.2 1B",
      "provider": "Meta",
      "params": 1,
      "active_params": null,
      "is_moe": false,
      "kv_head_fraction": 0.25,
      "arch": {
        "layers": 16,
        "hidden": 2048,
        "intermediate": 8192,
        "heads": 32,
        "kv_heads": 8,
        "head_dim": 64,
        "vocab": 128256,
        "tie_embeddings": true,
        "max_position": 131072
      },
      "ollama_tag": "llama3.2:1b",
      "hf_repo": "meta-llama/Llama-3.2-1B-Instruct",
      "modalities": [
        "text"
      ],
      "note": null,
      "release_date": "2024-09",
      "popularity": 67
    },
    {
      "id": "llama-3-2-3b",
      "name": "Llama 3.2 3B",
      "provider": "Meta",
      "params": 3.21,
      "active_params": null,
      "is_moe": false,
      "kv_head_fraction": 0.3333,
      "arch": {
        "layers": 28,
        "hidden": 3072,
        "intermediate": 8192,
        "heads": 24,
        "kv_heads": 8,
        "head_dim": 128,
        "vocab": 128256,
        "tie_embeddings": true,
        "max_position": 131072
      },
      "ollama_tag": "llama3.2:3b",
      "hf_repo": "meta-llama/Llama-3.2-3B-Instruct",
      "modalities": [
        "text"
      ],
      "note": null,
      "release_date": "2024-09",
      "popularity": 60
    },
    {
      "id": "llama-3-2-vision-11b",
      "name": "Llama 3.2 Vision 11B",
      "provider": "Meta",
      "params": 10.67,
      "active_params": null,
      "is_moe": false,
      "kv_head_fraction": 0.25,
      "arch": {
        "layers": 40,
        "hidden": 4096,
        "intermediate": 14336,
        "heads": 32,
        "kv_heads": 8,
        "vocab": 128256,
        "tie_embeddings": false,
        "max_position": 131072
      },
      "ollama_tag": "llama3.2-vision:11b",
      "hf_repo": "meta-llama/Llama-3.2-11B-Vision-Instruct",
      "modalities": [
        "text",
        "image"
      ],
      "note": "Multimodal (text + image)",
      "release_date": "2024-09",
      "popularity": 15
    },
    {
      "id": "llama-3-1-8b",
      "name": "Llama 3.1 8B",
      "provider": "Meta",
      "params": 8,
      "active_params": null,
      "is_moe": false,
      "kv_head_fraction": 0.25,
      "arch": {
        "layers": 32,
        "hidden": 4096,
        "intermediate": 14336,
        "heads": 32,
        "kv_heads": 8,
        "head_dim": 128,
        "vocab": 128256,
        "tie_embeddings": false,
        "max_position": 131072
      },
      "ollama_tag": "llama3.1:8b",
      "hf_repo": "meta-llama/Llama-3.1-8B-Instruct",
      "modalities": [
        "text"
      ],
      "note": "Grades set 2026-07-11 by peer comparison to same-era small dense models (llama-3-2-3b, qwen-2-5-7b) using the 5 published benchmarks; no agent/vision/multilingual benchmarks published for this row.",
      "release_date": "2024-07",
      "popularity": 86
    },
    {
      "id": "llama-3-3-70b",
      "name": "Llama 3.3 70B",
      "provider": "Meta",
      "params": 70,
      "active_params": null,
      "is_moe": false,
      "kv_head_fraction": 0.125,
      "arch": {
        "layers": 80,
        "hidden": 8192,
        "intermediate": 28672,
        "heads": 64,
        "kv_heads": 8,
        "head_dim": 128,
        "vocab": 128256,
        "tie_embeddings": false,
        "max_position": 131072
      },
      "ollama_tag": "llama3.3:70b",
      "hf_repo": "meta-llama/Llama-3.3-70B-Instruct",
      "modalities": [
        "text"
      ],
      "note": null,
      "release_date": "2024-12",
      "popularity": 47
    },
    {
      "id": "llama-3-2-vision-90b",
      "name": "Llama 3.2 Vision 90B",
      "provider": "Meta",
      "params": 90,
      "active_params": null,
      "is_moe": false,
      "kv_head_fraction": 0.125,
      "arch": {
        "layers": 100,
        "hidden": 8192,
        "intermediate": 28672,
        "heads": 64,
        "kv_heads": 8,
        "vocab": 128256,
        "tie_embeddings": false,
        "max_position": 131072
      },
      "ollama_tag": "llama3.2-vision:90b",
      "hf_repo": "meta-llama/Llama-3.2-90B-Vision-Instruct",
      "modalities": [
        "text",
        "image"
      ],
      "note": "Multimodal (text + image)",
      "release_date": "2024-09",
      "popularity": 0
    },
    {
      "id": "gemma-2-2b",
      "name": "Gemma 2 2B",
      "provider": "Google",
      "params": 2,
      "active_params": null,
      "is_moe": false,
      "kv_head_fraction": 0.5,
      "arch": {
        "layers": 26,
        "hidden": 2304,
        "intermediate": 9216,
        "heads": 8,
        "kv_heads": 4,
        "head_dim": 256,
        "vocab": 256000,
        "max_position": 8192,
        "sliding_window": 4096,
        "sliding_layers": 13
      },
      "ollama_tag": "gemma2:2b",
      "hf_repo": "google/gemma-2-2b-it",
      "modalities": [
        "text"
      ],
      "note": null,
      "release_date": "2024-07",
      "popularity": 32
    },
    {
      "id": "gemma-2-9b",
      "name": "Gemma 2 9B",
      "provider": "Google",
      "params": 9,
      "active_params": null,
      "is_moe": false,
      "kv_head_fraction": 0.5,
      "arch": {
        "layers": 42,
        "hidden": 3584,
        "intermediate": 14336,
        "heads": 16,
        "kv_heads": 8,
        "head_dim": 256,
        "vocab": 256000,
        "max_position": 8192,
        "sliding_window": 4096,
        "sliding_layers": 21
      },
      "ollama_tag": "gemma2:9b",
      "hf_repo": "google/gemma-2-9b-it",
      "modalities": [
        "text"
      ],
      "note": "Grades set 2026-07-11 by peer comparison to same-era small dense models; weaker than llama-3-1-8b on all 3 shared benchmarks (HumanEval, GSM8K, ARC-C) and has no MMLU-Pro/IFEval published. 8K context is also the smallest in the registry, limiting long-document/agent use.",
      "release_date": "2024-06",
      "popularity": 29
    },
    {
      "id": "gemma-2-27b",
      "name": "Gemma 2 27B",
      "provider": "Google",
      "params": 27,
      "active_params": null,
      "is_moe": false,
      "kv_head_fraction": 0.5,
      "arch": {
        "layers": 46,
        "hidden": 4608,
        "intermediate": 36864,
        "heads": 32,
        "kv_heads": 16,
        "head_dim": 128,
        "vocab": 256000,
        "max_position": 8192,
        "sliding_window": 4096,
        "sliding_layers": 23
      },
      "ollama_tag": "gemma2:27b",
      "hf_repo": "google/gemma-2-27b-it",
      "modalities": [
        "text"
      ],
      "note": null,
      "release_date": "2024-06",
      "popularity": 9
    },
    {
      "id": "gemma-3-1b",
      "name": "Gemma 3 1B",
      "provider": "Google",
      "params": 1,
      "active_params": null,
      "is_moe": false,
      "kv_head_fraction": 0.25,
      "arch": {
        "layers": 26,
        "hidden": 1152,
        "intermediate": 6912,
        "heads": 4,
        "kv_heads": 1,
        "head_dim": 256,
        "vocab": 262144,
        "max_position": 32768,
        "sliding_window": 512,
        "sliding_layers": 22
      },
      "ollama_tag": "gemma3:1b",
      "hf_repo": "google/gemma-3-1b-it",
      "modalities": [
        "text"
      ],
      "note": "Text-only, 32K ctx",
      "release_date": "2025-03",
      "popularity": 33
    },
    {
      "id": "gemma-3-4b",
      "name": "Gemma 3 4B",
      "provider": "Google",
      "params": 4.3,
      "active_params": null,
      "is_moe": false,
      "kv_head_fraction": 0.5,
      "arch": {
        "layers": 34,
        "hidden": 2560,
        "intermediate": 10240,
        "heads": 8,
        "kv_heads": 4,
        "head_dim": 256,
        "vocab": 262208,
        "max_position": 131072,
        "sliding_window": 1024,
        "sliding_layers": 29
      },
      "ollama_tag": "gemma3:4b",
      "hf_repo": "google/gemma-3-4b-it",
      "modalities": [
        "text",
        "image"
      ],
      "note": "Multimodal, 128K ctx",
      "release_date": "2025-03",
      "popularity": 61
    },
    {
      "id": "gemma-3-12b",
      "name": "Gemma 3 12B",
      "provider": "Google",
      "params": 12,
      "active_params": null,
      "is_moe": false,
      "kv_head_fraction": 0.5,
      "arch": {
        "layers": 48,
        "hidden": 3840,
        "intermediate": 15360,
        "heads": 16,
        "kv_heads": 8,
        "head_dim": 256,
        "vocab": 262208,
        "max_position": 131072,
        "sliding_window": 1024,
        "sliding_layers": 40
      },
      "ollama_tag": "gemma3:12b",
      "hf_repo": "google/gemma-3-12b-it",
      "modalities": [
        "text",
        "image"
      ],
      "note": "Multimodal, 128K ctx",
      "release_date": "2025-03",
      "popularity": 51
    },
    {
      "id": "gemma-3-27b",
      "name": "Gemma 3 27B",
      "provider": "Google",
      "params": 27,
      "active_params": null,
      "is_moe": false,
      "kv_head_fraction": 0.5,
      "arch": {
        "layers": 62,
        "hidden": 5376,
        "intermediate": 21504,
        "heads": 32,
        "kv_heads": 16,
        "head_dim": 128,
        "vocab": 262208,
        "max_position": 131072,
        "sliding_window": 1024,
        "sliding_layers": 52
      },
      "ollama_tag": "gemma3:27b",
      "hf_repo": "google/gemma-3-27b-it",
      "modalities": [
        "text",
        "image"
      ],
      "note": "Multimodal, 128K ctx",
      "release_date": "2025-03",
      "popularity": 41
    },
    {
      "id": "gemma-4-26b",
      "name": "Gemma 4 26B (MoE)",
      "provider": "Google",
      "params": 25.2,
      "active_params": 4,
      "is_moe": true,
      "kv_head_fraction": 0.5,
      "arch": {
        "layers": 30,
        "hidden": 2816,
        "intermediate": 2112,
        "heads": 16,
        "kv_heads": 8,
        "head_dim": 256,
        "vocab": 262144,
        "tie_embeddings": true,
        "max_position": 262144,
        "experts": 128,
        "moe_intermediate": 704,
        "sliding_window": 1024,
        "sliding_layers": 25
      },
      "ollama_tag": "gemma4:26b",
      "hf_repo": "google/gemma-4-26B-A4B-it",
      "modalities": [
        "text",
        "image"
      ],
      "note": "MoE — ~4B active of 25.2B total, 128 experts. Multimodal, 256K ctx",
      "release_date": "2026-04",
      "popularity": 90
    },
    {
      "id": "gemma-4-e2b",
      "name": "Gemma 4 E2B",
      "provider": "Google",
      "params": 2.3,
      "active_params": null,
      "is_moe": false,
      "kv_head_fraction": 0.125,
      "arch": {
        "layers": 35,
        "hidden": 1536,
        "intermediate": 6144,
        "heads": 8,
        "kv_heads": 1,
        "head_dim": 256,
        "vocab": 262144,
        "tie_embeddings": true,
        "max_position": 131072,
        "sliding_window": 512,
        "sliding_layers": 28
      },
      "ollama_tag": "gemma4:e2b",
      "hf_repo": "google/gemma-4-E2B-it",
      "modalities": [
        "text",
        "image"
      ],
      "note": "Edge model. Multimodal, 128K ctx",
      "release_date": "2026-03",
      "popularity": 81
    },
    {
      "id": "gemma-4-e4b",
      "name": "Gemma 4 E4B",
      "provider": "Google",
      "params": 4.5,
      "active_params": null,
      "is_moe": false,
      "kv_head_fraction": 0.25,
      "arch": {
        "layers": 42,
        "hidden": 2560,
        "intermediate": 10240,
        "heads": 8,
        "kv_heads": 2,
        "head_dim": 256,
        "vocab": 262144,
        "tie_embeddings": true,
        "max_position": 131072,
        "sliding_window": 512,
        "sliding_layers": 35
      },
      "ollama_tag": "gemma4:e4b",
      "hf_repo": "google/gemma-4-E4B-it",
      "modalities": [
        "text",
        "image"
      ],
      "note": "Edge model. Multimodal, 128K ctx",
      "release_date": "2026-03",
      "popularity": 87
    },
    {
      "id": "mistral-7b",
      "name": "Mistral 7B (v0.3)",
      "provider": "Mistral AI",
      "params": 7.25,
      "active_params": null,
      "is_moe": false,
      "kv_head_fraction": 0.25,
      "arch": {
        "layers": 32,
        "hidden": 4096,
        "intermediate": 14336,
        "heads": 32,
        "kv_heads": 8,
        "vocab": 32768,
        "tie_embeddings": false,
        "max_position": 32768
      },
      "ollama_tag": "mistral:7b",
      "hf_repo": "mistralai/Mistral-7B-Instruct-v0.3",
      "modalities": [
        "text"
      ],
      "note": null,
      "release_date": "2024-05",
      "popularity": 76
    },
    {
      "id": "mistral-nemo-12b",
      "name": "Mistral Nemo 12B",
      "provider": "Mistral AI",
      "params": 12,
      "active_params": null,
      "is_moe": false,
      "kv_head_fraction": 0.25,
      "arch": {
        "layers": 40,
        "hidden": 5120,
        "intermediate": 14336,
        "heads": 32,
        "kv_heads": 8,
        "head_dim": 128,
        "vocab": 131072,
        "tie_embeddings": false,
        "max_position": 131072
      },
      "ollama_tag": "mistral-nemo:12b",
      "hf_repo": "mistralai/Mistral-Nemo-Instruct-2407",
      "modalities": [
        "text"
      ],
      "note": null,
      "release_date": "2024-07",
      "popularity": 20
    },
    {
      "id": "mistral-small-3-1-24b",
      "name": "Mistral Small 3.1 24B",
      "provider": "Mistral AI",
      "params": 24,
      "active_params": null,
      "is_moe": false,
      "kv_head_fraction": 0.25,
      "arch": {
        "layers": 40,
        "hidden": 5120,
        "intermediate": 32768,
        "heads": 32,
        "kv_heads": 8,
        "head_dim": 128,
        "vocab": 131072,
        "max_position": 131072
      },
      "ollama_tag": "mistral-small:24b",
      "hf_repo": "mistralai/Mistral-Small-3.1-24B-Instruct-2503",
      "modalities": [
        "text",
        "image"
      ],
      "note": "Multimodal, 128K ctx",
      "release_date": "2025-03",
      "popularity": 28
    },
    {
      "id": "mistral-small-3-2-24b",
      "name": "Mistral Small 3.2 24B",
      "provider": "Mistral AI",
      "params": 24,
      "active_params": null,
      "is_moe": false,
      "kv_head_fraction": 0.25,
      "arch": {
        "layers": 40,
        "hidden": 5120,
        "intermediate": 32768,
        "heads": 32,
        "kv_heads": 8,
        "head_dim": 128,
        "vocab": 131072,
        "max_position": 131072
      },
      "ollama_tag": "mistral-small3.2:24b",
      "hf_repo": "mistralai/Mistral-Small-3.2-24B-Instruct-2506",
      "modalities": [
        "text",
        "image"
      ],
      "note": "Multimodal, 128K ctx. Improved instruction following vs 3.1",
      "release_date": "2025-06",
      "popularity": 22
    },
    {
      "id": "ministral-3b",
      "name": "Ministral 3B",
      "provider": "Mistral AI",
      "params": 3.8,
      "active_params": null,
      "is_moe": false,
      "kv_head_fraction": 0.25,
      "arch": {
        "layers": 26,
        "hidden": 3072,
        "intermediate": 9216,
        "heads": 32,
        "kv_heads": 8,
        "head_dim": 128,
        "vocab": 131072,
        "tie_embeddings": true,
        "max_position": 262144
      },
      "ollama_tag": "ministral-3:3b",
      "hf_repo": "mistralai/Ministral-3-3B-Instruct-2512",
      "modalities": [
        "text",
        "image"
      ],
      "note": "Dense. Vision, 256K ctx. Ministral 3 series, Apache 2.0.",
      "release_date": "2025-12",
      "popularity": 17
    },
    {
      "id": "ministral-8b",
      "name": "Ministral 8B",
      "provider": "Mistral AI",
      "params": 8.9,
      "active_params": null,
      "is_moe": false,
      "kv_head_fraction": 0.25,
      "arch": {
        "layers": 34,
        "hidden": 4096,
        "intermediate": 14336,
        "heads": 32,
        "kv_heads": 8,
        "head_dim": 128,
        "vocab": 131072,
        "tie_embeddings": false,
        "max_position": 262144
      },
      "ollama_tag": "ministral-3:8b",
      "hf_repo": "mistralai/Ministral-3-8B-Instruct-2512",
      "modalities": [
        "text",
        "image"
      ],
      "note": "Dense. Vision, 256K ctx. Ministral 3 series, Apache 2.0.",
      "release_date": "2025-12",
      "popularity": 21
    },
    {
      "id": "ministral-14b",
      "name": "Ministral 14B",
      "provider": "Mistral AI",
      "params": 13.9,
      "active_params": null,
      "is_moe": false,
      "kv_head_fraction": 0.25,
      "arch": {
        "layers": 40,
        "hidden": 5120,
        "intermediate": 16384,
        "heads": 32,
        "kv_heads": 8,
        "head_dim": 128,
        "vocab": 131072,
        "tie_embeddings": false,
        "max_position": 262144
      },
      "ollama_tag": "ministral-3:14b",
      "hf_repo": "mistralai/Ministral-3-14B-Instruct-2512",
      "modalities": [
        "text",
        "image"
      ],
      "note": "Dense. Vision, 256K ctx. Ministral 3 series, Apache 2.0.",
      "release_date": "2025-12",
      "popularity": 35
    },
    {
      "id": "mixtral-8x7b",
      "name": "Mixtral 8x7B",
      "provider": "Mistral AI",
      "params": 46.7,
      "active_params": 13.1,
      "is_moe": true,
      "kv_head_fraction": 0.25,
      "arch": {
        "layers": 32,
        "hidden": 4096,
        "intermediate": 14336,
        "heads": 32,
        "kv_heads": 8,
        "vocab": 32000,
        "tie_embeddings": false,
        "max_position": 32768,
        "experts": 8,
        "experts_per_tok": 2
      },
      "ollama_tag": "mixtral:8x7b",
      "hf_repo": "mistralai/Mixtral-8x7B-Instruct-v0.1",
      "modalities": [
        "text"
      ],
      "note": "MoE — ~13B active of 46.7B total",
      "release_date": "2023-12",
      "popularity": 46
    },
    {
      "id": "phi-3-mini-3-8b",
      "name": "Phi-3 Mini 3.8B",
      "provider": "Microsoft",
      "params": 3.8,
      "active_params": null,
      "is_moe": false,
      "kv_head_fraction": 1,
      "arch": {
        "layers": 32,
        "hidden": 3072,
        "intermediate": 8192,
        "heads": 32,
        "kv_heads": 32,
        "vocab": 32064,
        "tie_embeddings": false,
        "max_position": 4096
      },
      "ollama_tag": "phi3:3.8b",
      "hf_repo": "microsoft/Phi-3-mini-4k-instruct",
      "modalities": [
        "text"
      ],
      "note": null,
      "release_date": "2024-04",
      "popularity": 44
    },
    {
      "id": "phi-3-medium-14b",
      "name": "Phi-3 Medium 14B",
      "provider": "Microsoft",
      "params": 14,
      "active_params": null,
      "is_moe": false,
      "kv_head_fraction": 0.25,
      "arch": {
        "layers": 40,
        "hidden": 5120,
        "intermediate": 17920,
        "heads": 40,
        "kv_heads": 10,
        "vocab": 32064,
        "tie_embeddings": false,
        "max_position": 4096
      },
      "ollama_tag": "phi3:14b",
      "hf_repo": "microsoft/Phi-3-medium-4k-instruct",
      "modalities": [
        "text"
      ],
      "note": null,
      "release_date": "2024-05",
      "popularity": 11
    },
    {
      "id": "phi-4-14b",
      "name": "Phi-4 14B",
      "provider": "Microsoft",
      "params": 14.66,
      "active_params": null,
      "is_moe": false,
      "kv_head_fraction": 0.25,
      "arch": {
        "layers": 40,
        "hidden": 5120,
        "intermediate": 17920,
        "heads": 40,
        "kv_heads": 10,
        "vocab": 100352,
        "tie_embeddings": false,
        "max_position": 16384
      },
      "ollama_tag": "phi4:14b",
      "hf_repo": "microsoft/phi-4",
      "modalities": [
        "text"
      ],
      "note": null,
      "release_date": "2024-12",
      "popularity": 36
    },
    {
      "id": "phi-4-mini-3-8b",
      "name": "Phi-4 Mini 3.8B",
      "provider": "Microsoft",
      "params": 3.8,
      "active_params": null,
      "is_moe": false,
      "kv_head_fraction": 0.3333,
      "arch": {
        "layers": 32,
        "hidden": 3072,
        "intermediate": 8192,
        "heads": 24,
        "kv_heads": 8,
        "vocab": 200064,
        "tie_embeddings": true,
        "max_position": 131072
      },
      "ollama_tag": "phi4-mini:3.8b",
      "hf_repo": "microsoft/Phi-4-mini-instruct",
      "modalities": [
        "text"
      ],
      "note": "128K ctx, function calling",
      "release_date": "2025-02",
      "popularity": 39
    },
    {
      "id": "phi-4-reasoning-plus-14b",
      "name": "Phi-4-reasoning-plus 14B",
      "provider": "Microsoft",
      "params": 14.66,
      "active_params": null,
      "is_moe": false,
      "kv_head_fraction": 0.25,
      "arch": {
        "layers": 40,
        "hidden": 5120,
        "intermediate": 17920,
        "heads": 40,
        "kv_heads": 10,
        "vocab": 100352,
        "tie_embeddings": false,
        "max_position": 32768
      },
      "ollama_tag": "phi4-reasoning:plus",
      "hf_repo": "microsoft/Phi-4-reasoning-plus",
      "modalities": [
        "text"
      ],
      "note": "Reasoning model. 32K ctx. MIT license",
      "release_date": "2025-04",
      "popularity": 4
    },
    {
      "id": "qwen-2-5-0-5b",
      "name": "Qwen 2.5 0.5B",
      "provider": "Alibaba",
      "params": 0.5,
      "active_params": null,
      "is_moe": false,
      "kv_head_fraction": 0.1429,
      "arch": {
        "layers": 24,
        "hidden": 896,
        "intermediate": 4864,
        "heads": 14,
        "kv_heads": 2,
        "vocab": 151936,
        "tie_embeddings": true,
        "max_position": 32768
      },
      "ollama_tag": "qwen2.5:0.5b",
      "hf_repo": "Qwen/Qwen2.5-0.5B-Instruct",
      "modalities": [
        "text"
      ],
      "note": null,
      "release_date": "2024-09",
      "popularity": 85
    },
    {
      "id": "qwen-2-5-1-5b",
      "name": "Qwen 2.5 1.5B",
      "provider": "Alibaba",
      "params": 1.5,
      "active_params": null,
      "is_moe": false,
      "kv_head_fraction": 0.1667,
      "arch": {
        "layers": 28,
        "hidden": 1536,
        "intermediate": 8960,
        "heads": 12,
        "kv_heads": 2,
        "vocab": 151936,
        "tie_embeddings": true,
        "max_position": 32768
      },
      "ollama_tag": "qwen2.5:1.5b",
      "hf_repo": "Qwen/Qwen2.5-1.5B-Instruct",
      "modalities": [
        "text"
      ],
      "note": null,
      "release_date": "2024-09",
      "popularity": 89
    },
    {
      "id": "qwen-2-5-7b",
      "name": "Qwen 2.5 7B",
      "provider": "Alibaba",
      "params": 7.62,
      "active_params": null,
      "is_moe": false,
      "kv_head_fraction": 0.1429,
      "arch": {
        "layers": 28,
        "hidden": 3584,
        "intermediate": 18944,
        "heads": 28,
        "kv_heads": 4,
        "vocab": 152064,
        "tie_embeddings": false,
        "max_position": 32768
      },
      "ollama_tag": "qwen2.5:7b",
      "hf_repo": "Qwen/Qwen2.5-7B-Instruct",
      "modalities": [
        "text"
      ],
      "note": null,
      "release_date": "2024-09",
      "popularity": 96
    },
    {
      "id": "qwen-2-5-14b",
      "name": "Qwen 2.5 14B",
      "provider": "Alibaba",
      "params": 14.77,
      "active_params": null,
      "is_moe": false,
      "kv_head_fraction": 0.2,
      "arch": {
        "layers": 48,
        "hidden": 5120,
        "intermediate": 13824,
        "heads": 40,
        "kv_heads": 8,
        "vocab": 152064,
        "tie_embeddings": false,
        "max_position": 32768
      },
      "ollama_tag": "qwen2.5:14b",
      "hf_repo": "Qwen/Qwen2.5-14B-Instruct",
      "modalities": [
        "text"
      ],
      "note": null,
      "release_date": "2024-09",
      "popularity": 54
    },
    {
      "id": "qwen-2-5-32b",
      "name": "Qwen 2.5 32B",
      "provider": "Alibaba",
      "params": 32,
      "active_params": null,
      "is_moe": false,
      "kv_head_fraction": 0.2,
      "arch": {
        "layers": 64,
        "hidden": 5120,
        "intermediate": 27648,
        "heads": 40,
        "kv_heads": 8,
        "vocab": 152064,
        "tie_embeddings": false,
        "max_position": 32768
      },
      "ollama_tag": "qwen2.5:32b",
      "hf_repo": "Qwen/Qwen2.5-32B-Instruct",
      "modalities": [
        "text"
      ],
      "note": null,
      "release_date": "2024-09",
      "popularity": 55
    },
    {
      "id": "qwen-2-5-72b",
      "name": "Qwen 2.5 72B",
      "provider": "Alibaba",
      "params": 72,
      "active_params": null,
      "is_moe": false,
      "kv_head_fraction": 0.125,
      "arch": {
        "layers": 80,
        "hidden": 8192,
        "intermediate": 29568,
        "heads": 64,
        "kv_heads": 8,
        "vocab": 152064,
        "tie_embeddings": false,
        "max_position": 32768
      },
      "ollama_tag": "qwen2.5:72b",
      "hf_repo": "Qwen/Qwen2.5-72B-Instruct",
      "modalities": [
        "text"
      ],
      "note": null,
      "release_date": "2024-09",
      "popularity": 30
    },
    {
      "id": "qwen-2-5-coder-7b",
      "name": "Qwen 2.5-Coder 7B",
      "provider": "Alibaba",
      "params": 7.62,
      "active_params": null,
      "is_moe": false,
      "kv_head_fraction": 0.1429,
      "arch": {
        "layers": 28,
        "hidden": 3584,
        "intermediate": 18944,
        "heads": 28,
        "kv_heads": 4,
        "vocab": 152064,
        "tie_embeddings": false,
        "max_position": 32768
      },
      "ollama_tag": "qwen2.5-coder:7b",
      "hf_repo": "Qwen/Qwen2.5-Coder-7B-Instruct",
      "modalities": [
        "text"
      ],
      "note": null,
      "release_date": "2024-09",
      "popularity": 69
    },
    {
      "id": "qwen-2-5-coder-32b",
      "name": "Qwen 2.5-Coder 32B",
      "provider": "Alibaba",
      "params": 32,
      "active_params": null,
      "is_moe": false,
      "kv_head_fraction": 0.2,
      "arch": {
        "layers": 64,
        "hidden": 5120,
        "intermediate": 27648,
        "heads": 40,
        "kv_heads": 8,
        "vocab": 152064,
        "tie_embeddings": false,
        "max_position": 32768
      },
      "ollama_tag": "qwen2.5-coder:32b",
      "hf_repo": "Qwen/Qwen2.5-Coder-32B-Instruct",
      "modalities": [
        "text"
      ],
      "note": null,
      "release_date": "2024-11",
      "popularity": 53
    },
    {
      "id": "qwen-3-0-6b",
      "name": "Qwen 3 0.6B",
      "provider": "Alibaba",
      "params": 0.6,
      "active_params": null,
      "is_moe": false,
      "kv_head_fraction": 0.5,
      "arch": {
        "layers": 28,
        "hidden": 1024,
        "intermediate": 3072,
        "heads": 16,
        "kv_heads": 8,
        "head_dim": 128,
        "vocab": 151936,
        "tie_embeddings": true,
        "max_position": 40960
      },
      "ollama_tag": "qwen3:0.6b",
      "hf_repo": "Qwen/Qwen3-0.6B",
      "modalities": [
        "text"
      ],
      "note": "Dense",
      "release_date": "2025-04",
      "popularity": 100
    },
    {
      "id": "qwen-3-1-7b",
      "name": "Qwen 3 1.7B",
      "provider": "Alibaba",
      "params": 1.7,
      "active_params": null,
      "is_moe": false,
      "kv_head_fraction": 0.5,
      "arch": {
        "layers": 28,
        "hidden": 2048,
        "intermediate": 6144,
        "heads": 16,
        "kv_heads": 8,
        "head_dim": 128,
        "vocab": 151936,
        "tie_embeddings": true,
        "max_position": 40960
      },
      "ollama_tag": "qwen3:1.7b",
      "hf_repo": "Qwen/Qwen3-1.7B",
      "modalities": [
        "text"
      ],
      "note": "Dense",
      "release_date": "2025-04",
      "popularity": 79
    },
    {
      "id": "qwen-3-4b",
      "name": "Qwen 3 4B",
      "provider": "Alibaba",
      "params": 4,
      "active_params": null,
      "is_moe": false,
      "kv_head_fraction": 0.25,
      "arch": {
        "layers": 36,
        "hidden": 2560,
        "intermediate": 9728,
        "heads": 32,
        "kv_heads": 8,
        "head_dim": 128,
        "vocab": 151936,
        "tie_embeddings": true,
        "max_position": 40960
      },
      "ollama_tag": "qwen3:4b",
      "hf_repo": "Qwen/Qwen3-4B",
      "modalities": [
        "text"
      ],
      "note": "Dense",
      "release_date": "2025-04",
      "popularity": 82
    },
    {
      "id": "qwen-3-8b",
      "name": "Qwen 3 8B",
      "provider": "Alibaba",
      "params": 8,
      "active_params": null,
      "is_moe": false,
      "kv_head_fraction": 0.25,
      "arch": {
        "layers": 36,
        "hidden": 4096,
        "intermediate": 12288,
        "heads": 32,
        "kv_heads": 8,
        "head_dim": 128,
        "vocab": 151936,
        "tie_embeddings": false,
        "max_position": 40960
      },
      "ollama_tag": "qwen3:8b",
      "hf_repo": "Qwen/Qwen3-8B",
      "modalities": [
        "text"
      ],
      "note": "Dense",
      "release_date": "2025-04",
      "popularity": 92
    },
    {
      "id": "qwen-3-14b",
      "name": "Qwen 3 14B",
      "provider": "Alibaba",
      "params": 14.77,
      "active_params": null,
      "is_moe": false,
      "kv_head_fraction": 0.2,
      "arch": {
        "layers": 40,
        "hidden": 5120,
        "intermediate": 17408,
        "heads": 40,
        "kv_heads": 8,
        "head_dim": 128,
        "vocab": 151936,
        "tie_embeddings": false,
        "max_position": 40960
      },
      "ollama_tag": "qwen3:14b",
      "hf_repo": "Qwen/Qwen3-14B",
      "modalities": [
        "text"
      ],
      "note": "Dense",
      "release_date": "2025-04",
      "popularity": 99
    },
    {
      "id": "qwen-3-32b",
      "name": "Qwen 3 32B",
      "provider": "Alibaba",
      "params": 32,
      "active_params": null,
      "is_moe": false,
      "kv_head_fraction": 0.125,
      "arch": {
        "layers": 64,
        "hidden": 5120,
        "intermediate": 25600,
        "heads": 64,
        "kv_heads": 8,
        "head_dim": 128,
        "vocab": 151936,
        "tie_embeddings": false,
        "max_position": 40960
      },
      "ollama_tag": "qwen3:32b",
      "hf_repo": "Qwen/Qwen3-32B",
      "modalities": [
        "text"
      ],
      "note": "Dense",
      "release_date": "2025-04",
      "popularity": 83
    },
    {
      "id": "qwen-3-30b-a3b",
      "name": "Qwen 3 30B-A3B",
      "provider": "Alibaba",
      "params": 30,
      "active_params": 3,
      "is_moe": true,
      "kv_head_fraction": 0.125,
      "arch": {
        "layers": 48,
        "hidden": 2048,
        "intermediate": 6144,
        "heads": 32,
        "kv_heads": 4,
        "head_dim": 128,
        "vocab": 151936,
        "tie_embeddings": false,
        "max_position": 40960,
        "experts": 128,
        "experts_per_tok": 8,
        "moe_intermediate": 768
      },
      "ollama_tag": "qwen3:30b-a3b",
      "hf_repo": "Qwen/Qwen3-30B-A3B",
      "modalities": [
        "text"
      ],
      "note": "MoE — ~3B active of 30B total",
      "release_date": "2025-04",
      "popularity": 64
    },
    {
      "id": "qwen-3-235b-a22b",
      "name": "Qwen 3 235B-A22B",
      "provider": "Alibaba",
      "params": 235,
      "active_params": 22.1,
      "is_moe": true,
      "kv_head_fraction": 0.0625,
      "arch": {
        "layers": 94,
        "hidden": 4096,
        "intermediate": 12288,
        "heads": 64,
        "kv_heads": 4,
        "head_dim": 128,
        "vocab": 151936,
        "tie_embeddings": false,
        "max_position": 40960,
        "experts": 128,
        "experts_per_tok": 8,
        "moe_intermediate": 1536
      },
      "ollama_tag": "qwen3:235b-a22b",
      "hf_repo": "Qwen/Qwen3-235B-A22B",
      "modalities": [
        "text"
      ],
      "note": "MoE — ~22B active of 235B total",
      "release_date": "2025-04",
      "popularity": 43
    },
    {
      "id": "qwq-32b",
      "name": "QwQ 32B",
      "provider": "Alibaba",
      "params": 32.5,
      "active_params": null,
      "is_moe": false,
      "kv_head_fraction": 0.2,
      "arch": {
        "layers": 64,
        "hidden": 5120,
        "intermediate": 27648,
        "heads": 40,
        "kv_heads": 8,
        "vocab": 152064,
        "tie_embeddings": false,
        "max_position": 40960
      },
      "ollama_tag": "qwq:32b",
      "hf_repo": "Qwen/QwQ-32B",
      "modalities": [
        "text"
      ],
      "note": "Reasoning model",
      "release_date": "2025-03",
      "popularity": 13
    },
    {
      "id": "deepseek-r1-1-5b",
      "name": "DeepSeek R1 Distill 1.5B",
      "provider": "DeepSeek",
      "params": 1.5,
      "active_params": null,
      "is_moe": false,
      "kv_head_fraction": 0.1667,
      "arch": {
        "layers": 28,
        "hidden": 1536,
        "intermediate": 8960,
        "heads": 12,
        "kv_heads": 2,
        "vocab": 151936,
        "tie_embeddings": false,
        "max_position": 131072
      },
      "ollama_tag": "deepseek-r1:1.5b",
      "hf_repo": "deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B",
      "modalities": [
        "text"
      ],
      "note": null,
      "release_date": "2025-01",
      "popularity": 52
    },
    {
      "id": "deepseek-r1-7b",
      "name": "DeepSeek R1 Distill 7B",
      "provider": "DeepSeek",
      "params": 7.62,
      "active_params": null,
      "is_moe": false,
      "kv_head_fraction": 0.1429,
      "arch": {
        "layers": 28,
        "hidden": 3584,
        "intermediate": 18944,
        "heads": 28,
        "kv_heads": 4,
        "vocab": 152064,
        "tie_embeddings": false,
        "max_position": 131072
      },
      "ollama_tag": "deepseek-r1:7b",
      "hf_repo": "deepseek-ai/DeepSeek-R1-Distill-Qwen-7B",
      "modalities": [
        "text"
      ],
      "note": null,
      "release_date": "2025-01",
      "popularity": 38
    },
    {
      "id": "deepseek-r1-8b",
      "name": "DeepSeek R1 Distill 8B",
      "provider": "DeepSeek",
      "params": 8,
      "active_params": null,
      "is_moe": false,
      "kv_head_fraction": 0.25,
      "arch": {
        "layers": 32,
        "hidden": 4096,
        "intermediate": 14336,
        "heads": 32,
        "kv_heads": 8,
        "vocab": 128256,
        "tie_embeddings": false,
        "max_position": 131072
      },
      "ollama_tag": "deepseek-r1:8b",
      "hf_repo": "deepseek-ai/DeepSeek-R1-Distill-Llama-8B",
      "modalities": [
        "text"
      ],
      "note": null,
      "release_date": "2025-01",
      "popularity": 18
    },
    {
      "id": "deepseek-r1-14b",
      "name": "DeepSeek R1 Distill 14B",
      "provider": "DeepSeek",
      "params": 14.77,
      "active_params": null,
      "is_moe": false,
      "kv_head_fraction": 0.2,
      "arch": {
        "layers": 48,
        "hidden": 5120,
        "intermediate": 13824,
        "heads": 40,
        "kv_heads": 8,
        "vocab": 152064,
        "tie_embeddings": false,
        "max_position": 131072
      },
      "ollama_tag": "deepseek-r1:14b",
      "hf_repo": "deepseek-ai/DeepSeek-R1-Distill-Qwen-14B",
      "modalities": [
        "text"
      ],
      "note": null,
      "release_date": "2025-01",
      "popularity": 24
    },
    {
      "id": "deepseek-r1-32b",
      "name": "DeepSeek R1 Distill 32B",
      "provider": "DeepSeek",
      "params": 32,
      "active_params": null,
      "is_moe": false,
      "kv_head_fraction": 0.2,
      "arch": {
        "layers": 64,
        "hidden": 5120,
        "intermediate": 27648,
        "heads": 40,
        "kv_heads": 8,
        "vocab": 152064,
        "tie_embeddings": false,
        "max_position": 131072
      },
      "ollama_tag": "deepseek-r1:32b",
      "hf_repo": "deepseek-ai/DeepSeek-R1-Distill-Qwen-32B",
      "modalities": [
        "text"
      ],
      "note": null,
      "release_date": "2025-01",
      "popularity": 49
    },
    {
      "id": "deepseek-r1-70b",
      "name": "DeepSeek R1 Distill 70B",
      "provider": "DeepSeek",
      "params": 70,
      "active_params": null,
      "is_moe": false,
      "kv_head_fraction": 0.125,
      "arch": {
        "layers": 80,
        "hidden": 8192,
        "intermediate": 28672,
        "heads": 64,
        "kv_heads": 8,
        "head_dim": 128,
        "vocab": 128256,
        "tie_embeddings": false,
        "max_position": 131072
      },
      "ollama_tag": "deepseek-r1:70b",
      "hf_repo": "deepseek-ai/DeepSeek-R1-Distill-Llama-70B",
      "modalities": [
        "text"
      ],
      "note": null,
      "release_date": "2025-01",
      "popularity": 10
    },
    {
      "id": "deepseek-r1-671b",
      "name": "DeepSeek R1 (Full)",
      "provider": "DeepSeek",
      "params": 671,
      "active_params": 36.9,
      "is_moe": true,
      "kv_head_fraction": 0.07,
      "arch": {
        "layers": 61,
        "hidden": 7168,
        "intermediate": 18432,
        "heads": 128,
        "kv_heads": 128,
        "vocab": 129280,
        "tie_embeddings": false,
        "max_position": 163840,
        "kv_lora_rank": 512,
        "qk_rope_head_dim": 64,
        "experts": 256,
        "experts_per_tok": 8,
        "moe_intermediate": 2048,
        "shared_experts": 1
      },
      "ollama_tag": "deepseek-r1:671b",
      "hf_repo": "deepseek-ai/DeepSeek-R1",
      "modalities": [
        "text"
      ],
      "note": "MoE — ~37B active of 671B total. Uses MLA (not GQA)",
      "release_date": "2025-01",
      "popularity": 73
    },
    {
      "id": "command-r7b",
      "name": "Command R7B",
      "provider": "Cohere",
      "params": 7,
      "active_params": null,
      "is_moe": false,
      "kv_head_fraction": 0.25,
      "ollama_tag": "command-r7b:7b",
      "hf_repo": "CohereForAI/c4ai-command-r7b-12-2024",
      "modalities": [
        "text"
      ],
      "note": "RAG + tool use",
      "release_date": "2024-12",
      "popularity": 1
    },
    {
      "id": "command-r-35b",
      "name": "Command R 35B",
      "provider": "Cohere",
      "params": 35,
      "active_params": null,
      "is_moe": false,
      "kv_head_fraction": 0.125,
      "ollama_tag": "command-r:35b",
      "hf_repo": "CohereForAI/c4ai-command-r-v01",
      "modalities": [
        "text"
      ],
      "note": "RAG-optimized, 128K ctx",
      "release_date": "2024-03",
      "popularity": 27
    },
    {
      "id": "zaya1-8b",
      "name": "ZAYA1-8B",
      "provider": "Zyphra",
      "params": 8.84,
      "active_params": 0.76,
      "is_moe": true,
      "kv_head_fraction": 0.25,
      "arch": {
        "layers": 40,
        "hidden": 2048,
        "heads": 8,
        "kv_heads": 2,
        "head_dim": 128,
        "vocab": 262272,
        "tie_embeddings": true,
        "max_position": 131072,
        "experts": 16,
        "experts_per_tok": 1,
        "moe_intermediate": 2048,
        "approx": true
      },
      "ollama_tag": null,
      "hf_repo": "Zyphra/ZAYA1-8B",
      "modalities": [
        "text"
      ],
      "note": "Reasoning MoE — 8.4B total / 0.76B active, Apache 2.0, trained on AMD. Self-reported evals (Zyphra harness).",
      "release_date": "2026-05",
      "popularity": 6
    },
    {
      "id": "kimi-k2-7-code",
      "name": "Kimi K2.7-Code",
      "provider": "Moonshot AI",
      "params": 1000,
      "active_params": 32,
      "is_moe": true,
      "kv_head_fraction": null,
      "arch": {
        "layers": 61,
        "hidden": 7168,
        "intermediate": 18432,
        "heads": 64,
        "kv_heads": 64,
        "vocab": 163840,
        "tie_embeddings": false,
        "max_position": 262144,
        "kv_lora_rank": 512,
        "qk_rope_head_dim": 64,
        "experts": 384,
        "experts_per_tok": 8,
        "moe_intermediate": 2048,
        "shared_experts": 1
      },
      "ollama_tag": null,
      "hf_repo": "moonshotai/Kimi-K2.7-Code",
      "modalities": [
        "text",
        "image"
      ],
      "note": "Open-weight coding-focused multimodal model. Per release announcement: 256K context, ~30% lower reasoning-token usage. Also available via paid API (model id kimi-k2.7-code) -- corrected from self-hosted-only 2026-08-09; api_available was wrong. A same-weights 'HighSpeed' serving variant (model id kimi-k2.7-code-highspeed, ~180 tok/s, 2x price: $1.90/$8.00) exists on the vendor platform -- not added as a separate canonical row since it's an inference-serving tier of this same model, not a distinct architecture/weights release.",
      "release_date": "2026-06",
      "popularity": 45
    },
    {
      "id": "gemma-4-12b",
      "name": "Gemma 4 12B",
      "provider": "Google",
      "params": 12,
      "active_params": null,
      "is_moe": false,
      "kv_head_fraction": 0.5,
      "arch": {
        "layers": 48,
        "hidden": 3840,
        "intermediate": 15360,
        "heads": 16,
        "kv_heads": 8,
        "head_dim": 256,
        "vocab": 262144,
        "tie_embeddings": true,
        "max_position": 262144,
        "sliding_window": 1024,
        "sliding_layers": 40
      },
      "ollama_tag": null,
      "hf_repo": "google/gemma-4-12B-it",
      "modalities": [
        "text",
        "image"
      ],
      "note": "Open-weight Gemma 4 family addition. Multimodal (any-to-any pipeline).",
      "release_date": "2026-05",
      "popularity": 71
    },
    {
      "id": "qwen3-coder-next",
      "name": "Qwen3 Coder Next",
      "provider": "Alibaba",
      "params": 80,
      "active_params": 3,
      "is_moe": true,
      "kv_head_fraction": 0.125,
      "arch": {
        "layers": 48,
        "hidden": 2048,
        "intermediate": 5120,
        "heads": 16,
        "kv_heads": 2,
        "head_dim": 256,
        "vocab": 151936,
        "tie_embeddings": false,
        "max_position": 262144,
        "experts": 512,
        "experts_per_tok": 10,
        "moe_intermediate": 512,
        "shared_experts": 1,
        "linear_layers": 36
      },
      "ollama_tag": null,
      "hf_repo": "Qwen/Qwen3-Coder-Next",
      "modalities": [
        "text"
      ],
      "note": "Open-weight Qwen3 coding-specialized model. Apache 2.0 license.",
      "release_date": "2026-02",
      "popularity": 48
    },
    {
      "id": "minimax-m2-7",
      "name": "MiniMax M2.7",
      "provider": "MiniMax",
      "params": 230,
      "active_params": 10,
      "is_moe": true,
      "kv_head_fraction": 0.1667,
      "arch": {
        "layers": 62,
        "hidden": 3072,
        "intermediate": 1536,
        "heads": 48,
        "kv_heads": 8,
        "head_dim": 128,
        "vocab": 200064,
        "tie_embeddings": false,
        "max_position": 204800,
        "experts": 256,
        "experts_per_tok": 8
      },
      "ollama_tag": null,
      "hf_repo": "MiniMaxAI/MiniMax-M2.7",
      "modalities": [
        "text"
      ],
      "note": "Successor to MiniMax M2.5. Open-weight MoE — 230B total / 10B active, 200K context.",
      "release_date": "2026-04",
      "popularity": 37
    },
    {
      "id": "nemotron-3-ultra-550b",
      "name": "Nemotron 3 Ultra 550B A55B",
      "provider": "NVIDIA",
      "params": 550,
      "active_params": 55,
      "is_moe": true,
      "kv_head_fraction": 0.0313,
      "arch": {
        "layers": 108,
        "hidden": 8192,
        "intermediate": 5120,
        "heads": 64,
        "kv_heads": 2,
        "head_dim": 128,
        "vocab": 131072,
        "tie_embeddings": false,
        "max_position": 262144,
        "experts": 512,
        "experts_per_tok": 22,
        "moe_intermediate": 5120,
        "shared_experts": 1,
        "linear_layers": 96
      },
      "ollama_tag": null,
      "hf_repo": "nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16",
      "modalities": [
        "text"
      ],
      "note": "MoE (LatentMoE + Mamba-2 hybrid) — 550B total, 55B active. OpenMDW 1.1 license. 256K native context, up to 1M via vLLM flag.",
      "release_date": "2026-06",
      "popularity": 63
    },
    {
      "id": "glm-5-2",
      "name": "GLM-5.2",
      "provider": "Zhipu AI",
      "params": 753,
      "active_params": 40,
      "is_moe": true,
      "kv_head_fraction": null,
      "arch": {
        "layers": 78,
        "hidden": 6144,
        "intermediate": 12288,
        "heads": 64,
        "kv_heads": 64,
        "head_dim": 192,
        "vocab": 154880,
        "tie_embeddings": false,
        "max_position": 1048576,
        "kv_lora_rank": 512,
        "qk_rope_head_dim": 64,
        "experts": 256,
        "experts_per_tok": 8,
        "moe_intermediate": 2048,
        "shared_experts": 1
      },
      "ollama_tag": null,
      "hf_repo": "zai-org/GLM-5.2",
      "modalities": [
        "text"
      ],
      "note": "MIT-licensed open-weight MoE, 753B total / ~40B active across 384 experts. Coding-focused: SWE-bench Pro 62.1, FrontierSWE 74.4, HLE 40.5 (schema has no key for these).",
      "release_date": "2026-06",
      "popularity": 40
    },
    {
      "id": "minimax-m3",
      "name": "MiniMax M3",
      "provider": "MiniMax",
      "params": 428,
      "active_params": 23,
      "is_moe": true,
      "kv_head_fraction": null,
      "arch": {
        "layers": 60,
        "hidden": 6144,
        "intermediate": 3072,
        "heads": 64,
        "kv_heads": 4,
        "head_dim": 128,
        "vocab": 200064,
        "tie_embeddings": false,
        "max_position": 1048576,
        "experts": 128,
        "experts_per_tok": 4,
        "shared_experts": 1,
        "approx": true
      },
      "ollama_tag": null,
      "hf_repo": "MiniMaxAI/MiniMax-M3",
      "modalities": [
        "text",
        "image",
        "video"
      ],
      "note": "Open-weight 428B/23B-active MoE with MiniMax Sparse Attention for 1M context. gpqa/mmlu_pro/livecode are vendor-reported at launch; SWE-bench Verified 80.5 from model card. Corrected 2026-08-09: modalities was text-only, HF card confirms native text/image/video multimodality. Pricing re-confirmed unchanged ($0.30/$1.20 standard tier, <=512K input) via official platform.minimax.io docs -- an aggregator claim of a $0.24 discounted input rate was not corroborated by the vendor's own pricing page, so not adopted.",
      "release_date": "2026-06",
      "popularity": 26
    },
    {
      "id": "deepseek-v4-pro",
      "name": "DeepSeek V4-Pro",
      "provider": "DeepSeek",
      "params": 1600,
      "active_params": 49,
      "is_moe": true,
      "kv_head_fraction": null,
      "arch": {
        "layers": 61,
        "hidden": 7168,
        "heads": 128,
        "kv_heads": 1,
        "head_dim": 512,
        "vocab": 129280,
        "tie_embeddings": false,
        "max_position": 1048576,
        "experts": 384,
        "experts_per_tok": 6,
        "moe_intermediate": 3072,
        "shared_experts": 1,
        "approx": true
      },
      "ollama_tag": null,
      "hf_repo": "deepseek-ai/DeepSeek-V4-Pro",
      "modalities": [
        "text"
      ],
      "note": "MIT open-weight MoE, 1.6T total / 49B active. Flagship of the V4 family; 1M context. Weights are open but 1.6T total makes local hosting impractical (multi-GPU/cloud only).",
      "release_date": "2026-04",
      "popularity": 42
    },
    {
      "id": "deepseek-v4-flash",
      "name": "DeepSeek V4-Flash",
      "provider": "DeepSeek",
      "params": 284,
      "active_params": 13,
      "is_moe": true,
      "kv_head_fraction": null,
      "arch": {
        "layers": 43,
        "hidden": 4096,
        "heads": 64,
        "kv_heads": 1,
        "head_dim": 512,
        "vocab": 129280,
        "tie_embeddings": false,
        "max_position": 1048576,
        "experts": 256,
        "experts_per_tok": 6,
        "moe_intermediate": 2048,
        "shared_experts": 1,
        "approx": true
      },
      "ollama_tag": null,
      "hf_repo": "deepseek-ai/DeepSeek-V4-Flash",
      "modalities": [
        "text"
      ],
      "note": "MIT open-weight MoE, 284B total / 13B active. Efficient V4-family sibling with 1M context.",
      "release_date": "2026-04",
      "popularity": 66
    },
    {
      "id": "qwen3-5-9b",
      "name": "Qwen3.5 9B",
      "provider": "Alibaba",
      "params": 9.65,
      "active_params": null,
      "is_moe": false,
      "kv_head_fraction": 0.25,
      "arch": {
        "layers": 32,
        "hidden": 4096,
        "intermediate": 12288,
        "heads": 16,
        "kv_heads": 4,
        "head_dim": 256,
        "vocab": 248320,
        "tie_embeddings": false,
        "max_position": 262144,
        "linear_layers": 24
      },
      "ollama_tag": null,
      "hf_repo": "Qwen/Qwen3.5-9B",
      "modalities": [
        "text"
      ],
      "note": "Apache-2.0 dense 9B with strong reasoning for its size; 256K native context (extensible). Very self-host-friendly.",
      "release_date": "2026-02",
      "popularity": 93
    },
    {
      "id": "kimi-k2-6",
      "name": "Kimi K2.6",
      "provider": "Moonshot AI",
      "params": 1000,
      "active_params": 32,
      "is_moe": true,
      "kv_head_fraction": null,
      "arch": {
        "layers": 61,
        "hidden": 7168,
        "intermediate": 18432,
        "heads": 64,
        "kv_heads": 64,
        "vocab": 163840,
        "tie_embeddings": false,
        "max_position": 262144,
        "kv_lora_rank": 512,
        "qk_rope_head_dim": 64,
        "experts": 384,
        "experts_per_tok": 8,
        "moe_intermediate": 2048,
        "shared_experts": 1
      },
      "ollama_tag": null,
      "hf_repo": "moonshotai/Kimi-K2.6",
      "modalities": [
        "text",
        "image"
      ],
      "note": "Modified-MIT open-weight native-multimodal MoE, 1T total / 32B active (384 experts, 8 active) with a 400M vision encoder. 256K context.",
      "release_date": "2026-04",
      "popularity": 34
    },
    {
      "id": "qwen3-5-4b",
      "name": "Qwen3.5 4B",
      "provider": "Alibaba",
      "params": 4,
      "active_params": null,
      "is_moe": false,
      "kv_head_fraction": 0.25,
      "arch": {
        "layers": 32,
        "hidden": 2560,
        "intermediate": 9216,
        "heads": 16,
        "kv_heads": 4,
        "head_dim": 256,
        "vocab": 248320,
        "tie_embeddings": true,
        "max_position": 262144,
        "linear_layers": 24
      },
      "ollama_tag": "qwen3.5:4b",
      "hf_repo": "Qwen/Qwen3.5-4B",
      "modalities": [
        "text",
        "image"
      ],
      "note": "Apache-2.0 dense 4B, vision-capable, 256K native context. Small edge-friendly sibling of the Qwen3.5 family; not offered via Alibaba Cloud Model Studio API at audit time (self-host only).",
      "release_date": "2026-02",
      "popularity": 88
    },
    {
      "id": "qwen3-5-0-8b",
      "name": "Qwen3.5 0.8B",
      "provider": "Alibaba",
      "params": 0.87,
      "active_params": null,
      "is_moe": false,
      "kv_head_fraction": 0.25,
      "arch": {
        "layers": 24,
        "hidden": 1024,
        "intermediate": 3584,
        "heads": 8,
        "kv_heads": 2,
        "head_dim": 256,
        "vocab": 248320,
        "tie_embeddings": true,
        "max_position": 262144,
        "linear_layers": 18
      },
      "ollama_tag": "qwen3.5:0.8b",
      "hf_repo": "Qwen/Qwen3.5-0.8B",
      "modalities": [
        "text",
        "image"
      ],
      "note": "Apache-2.0 dense 0.8B, vision-capable, 256K native context. Smallest Qwen3.5 size; targets on-device/edge use. No clean instruct-specific benchmark table found at audit time (left null per honest-null policy).",
      "release_date": "2026-02",
      "popularity": 80
    },
    {
      "id": "qwen3-5-2b",
      "name": "Qwen3.5 2B",
      "provider": "Alibaba",
      "params": 2.27,
      "active_params": null,
      "is_moe": false,
      "kv_head_fraction": 0.25,
      "arch": {
        "layers": 24,
        "hidden": 2048,
        "intermediate": 6144,
        "heads": 8,
        "kv_heads": 2,
        "head_dim": 256,
        "vocab": 248320,
        "tie_embeddings": true,
        "max_position": 262144,
        "linear_layers": 18
      },
      "ollama_tag": "qwen3.5:2b",
      "hf_repo": "Qwen/Qwen3.5-2B",
      "modalities": [
        "text",
        "image"
      ],
      "note": "Apache-2.0 dense 2B, vision-capable, 256K native context. No clean instruct-specific benchmark table found at audit time (left null per honest-null policy).",
      "release_date": "2026-02",
      "popularity": 91
    },
    {
      "id": "qwen3-5-122b-a10b",
      "name": "Qwen3.5 122B-A10B",
      "provider": "Alibaba",
      "params": 122,
      "active_params": 10,
      "is_moe": true,
      "kv_head_fraction": 0.0625,
      "arch": {
        "layers": 48,
        "hidden": 3072,
        "heads": 32,
        "kv_heads": 2,
        "head_dim": 256,
        "vocab": 248320,
        "tie_embeddings": false,
        "max_position": 262144,
        "experts": 256,
        "experts_per_tok": 8,
        "moe_intermediate": 1024,
        "shared_experts": 1,
        "linear_layers": 36
      },
      "ollama_tag": "qwen3.5:122b",
      "hf_repo": "Qwen/Qwen3.5-122B-A10B",
      "modalities": [
        "text",
        "image"
      ],
      "note": "Apache-2.0 MoE (256 experts, top-8), 122B total / 10B active, vision-capable, 256K native context. Mid-large member of the Qwen3.5 family between 35B-A3B and the 397B-A17B flagship. Benchmarks filled 2026-07-12 from the official HF model card; scores confirm the peer-comparison grades set at add time (no change needed).",
      "release_date": "2026-02",
      "popularity": 59
    },
    {
      "id": "qwen3-6-27b",
      "name": "Qwen3.6 27B",
      "provider": "Alibaba",
      "params": 27,
      "active_params": null,
      "is_moe": false,
      "kv_head_fraction": 0.1667,
      "arch": {
        "layers": 64,
        "hidden": 5120,
        "intermediate": 17408,
        "heads": 24,
        "kv_heads": 4,
        "head_dim": 256,
        "vocab": 248320,
        "tie_embeddings": false,
        "max_position": 262144,
        "linear_layers": 48
      },
      "ollama_tag": "qwen3.6:27b",
      "hf_repo": "Qwen/Qwen3.6-27B",
      "modalities": [
        "text",
        "image"
      ],
      "note": "Apache-2.0 dense 27B, successor to Qwen3.5-27B (backlog candidate superseded — this row added instead per the GLM-5.1/5.2 precedent). Reported to outperform the 397B-A17B MoE flagship on agentic coding benchmarks: Terminal-Bench 2.0 59.3 (vendor: \"matches Claude 4.5 Opus\"), QwenWebBench 1487. Not offered via Alibaba Cloud Model Studio API at audit time (self-host only).",
      "release_date": "2026-04",
      "popularity": 78
    },
    {
      "id": "qwen3-6-35b-a3b",
      "name": "Qwen3.6 35B-A3B",
      "provider": "Alibaba",
      "params": 35,
      "active_params": 3,
      "is_moe": true,
      "kv_head_fraction": 0.125,
      "arch": {
        "layers": 40,
        "hidden": 2048,
        "heads": 16,
        "kv_heads": 2,
        "head_dim": 256,
        "vocab": 248320,
        "tie_embeddings": false,
        "max_position": 262144,
        "experts": 256,
        "experts_per_tok": 8,
        "moe_intermediate": 512,
        "shared_experts": 1,
        "linear_layers": 30
      },
      "ollama_tag": "qwen3.6:35b",
      "hf_repo": "Qwen/Qwen3.6-35B-A3B",
      "modalities": [
        "text",
        "image"
      ],
      "note": "Apache-2.0 MoE (3B active / 35B total), successor to Qwen3.5-35B-A3B (backlog candidate superseded — this row added instead per the GLM-5.1/5.2 precedent). API served as qwen3.6-flash on Alibaba Cloud Model Studio. Vendor reports it beats the dense Qwen3.5-27B on several coding benchmarks despite far fewer active params.",
      "release_date": "2026-04",
      "popularity": 84
    },
    {
      "id": "kimi-k3",
      "name": "Kimi K3",
      "provider": "Moonshot AI",
      "params": 2800,
      "active_params": 104,
      "is_moe": true,
      "kv_head_fraction": 1,
      "arch": {
        "layers": 93,
        "hidden": 7168,
        "intermediate": 33792,
        "heads": 96,
        "kv_heads": 96,
        "vocab": 163840,
        "tie_embeddings": false,
        "max_position": 1048576,
        "kv_lora_rank": 512,
        "qk_rope_head_dim": 64,
        "experts": 896,
        "experts_per_tok": 16,
        "moe_intermediate": 3072,
        "shared_experts": 2,
        "approx": true
      },
      "ollama_tag": null,
      "hf_repo": "moonshotai/Kimi-K3",
      "modalities": [
        "text",
        "image",
        "video"
      ],
      "note": "2.8T-param MoE (896 experts, 16 active + 2 shared, ~104B active per vendor card), Kimi Delta Attention hybrid linear-attention architecture, native multimodal (text/image/video), 1M context. Announced 2026-07-16, weights released 2026-07-26/27. Active-params figure conflicts across sources (vendor HF README states 104B explicitly; at least one third-party writeup estimates ~50B) -- used the vendor's own model-card statement per sourcing precedence. Legacy 20-benchmark suite mostly unpublished (vendor reports Terminal-Bench 2.1, GDPval-AA v2, BrowseComp, DeepSWE instead); only gpqa_diamond mapped cleanly to our schema. chatbot_arena skipped in canonical: arena.ai lists only a preliminary 'kimi-k3-max' variant (1485, Preliminary tag), not a plain Kimi K3 entry -- left null pending a non-preliminary, non-variant-ambiguous score.",
      "release_date": "2026-07",
      "popularity": 58
    },
    {
      "id": "qwen3-8-27b",
      "name": "Qwen3.8-27B",
      "provider": "Alibaba",
      "params": 27,
      "active_params": null,
      "is_moe": false,
      "kv_head_fraction": 0.1667,
      "arch": {
        "layers": 64,
        "hidden": 5120,
        "intermediate": 17408,
        "heads": 24,
        "kv_heads": 4,
        "head_dim": 256,
        "vocab": 248320,
        "tie_embeddings": false,
        "max_position": 262144,
        "linear_layers": 48
      },
      "ollama_tag": null,
      "hf_repo": "Qwen/Qwen3.8-27B",
      "modalities": [
        "text",
        "image",
        "video"
      ],
      "note": "Apache-2.0 dense 27B (28B with vision encoder), hybrid Gated-DeltaNet + full-attention. Native multimodal (text+image+video). 262K native context, 1M via YaRN.",
      "release_date": "2026-08",
      "popularity": 97
    },
    {
      "id": "glm-5-3-flash",
      "name": "GLM-5.3 Flash",
      "provider": "Zhipu AI",
      "params": 320,
      "active_params": 18,
      "is_moe": true,
      "kv_head_fraction": null,
      "arch": {
        "layers": 45,
        "hidden": 4096,
        "intermediate": 12288,
        "heads": 64,
        "kv_heads": 64,
        "head_dim": 256,
        "vocab": 154880,
        "tie_embeddings": false,
        "max_position": 1048576,
        "kv_lora_rank": 512,
        "qk_rope_head_dim": 0,
        "experts": 288,
        "experts_per_tok": 8,
        "moe_intermediate": 2048,
        "shared_experts": 1,
        "linear_layers": 34,
        "approx": true
      },
      "ollama_tag": null,
      "hf_repo": "zai-org/GLM-5.3-Flash",
      "modalities": [
        "text"
      ],
      "note": "Sparse MoE (320B total / 18B active). MIT license. Hybrid sparse+linear attention. Codename 'Ox Alpha'. DeepSWE v1.1: 63.4, AutomationBench: 48.8. Approaches Claude Opus 4.8 performance in coding at fraction of cost.",
      "release_date": "2026-08",
      "popularity": 95
    },
    {
      "id": "deepseek-v4-1-flash",
      "name": "DeepSeek V4.1 Flash",
      "provider": "DeepSeek",
      "params": 552,
      "active_params": 16,
      "is_moe": true,
      "kv_head_fraction": null,
      "arch": {
        "layers": 40,
        "hidden": 5120,
        "heads": 64,
        "kv_heads": 1,
        "head_dim": 512,
        "vocab": 129280,
        "tie_embeddings": false,
        "max_position": 1048576,
        "experts": 384,
        "experts_per_tok": 6,
        "moe_intermediate": 2304,
        "shared_experts": 1,
        "approx": true
      },
      "ollama_tag": null,
      "hf_repo": "deepseek-ai/DeepSeek-V4.1-Flash",
      "modalities": [
        "text",
        "image"
      ],
      "note": "Multimodal MoE optimized for agentic workloads and long-context processing at low cost.",
      "release_date": "2026-09",
      "popularity": 70
    },
    {
      "id": "mimo-v2-6-pro",
      "name": "MiMo-V2.6-Pro",
      "provider": "Xiaomi",
      "params": 1024.2,
      "active_params": 42,
      "is_moe": true,
      "kv_head_fraction": 0.0625,
      "arch": {
        "layers": 70,
        "hidden": 6144,
        "intermediate": 16384,
        "heads": 128,
        "kv_heads": 8,
        "head_dim": 192,
        "vocab": 152576,
        "tie_embeddings": false,
        "max_position": 1048576,
        "experts": 384,
        "experts_per_tok": 8,
        "moe_intermediate": 2048,
        "sliding_window": 128,
        "sliding_layers": 60
      },
      "ollama_tag": null,
      "hf_repo": "XiaomiMiMo/MiMo-V2.6-Pro-RL",
      "modalities": [
        "text",
        "image",
        "audio",
        "video"
      ],
      "note": "MiMo-V2.6-Pro is the flagship foundation model developed by Xiaomi.",
      "release_date": "2026-09",
      "popularity": 12
    },
    {
      "id": "mimo-v2-5",
      "name": "MiMo-V2.5",
      "provider": "Xiaomi",
      "params": 310.8,
      "active_params": 15,
      "is_moe": true,
      "kv_head_fraction": 0.0625,
      "arch": {
        "layers": 48,
        "hidden": 4096,
        "intermediate": 16384,
        "heads": 64,
        "kv_heads": 4,
        "head_dim": 192,
        "vocab": 152576,
        "tie_embeddings": false,
        "max_position": 1048576,
        "experts": 256,
        "experts_per_tok": 8,
        "moe_intermediate": 2048,
        "sliding_window": 128,
        "sliding_layers": 39
      },
      "ollama_tag": null,
      "hf_repo": "XiaomiMiMo/MiMo-V2.5",
      "modalities": [
        "text",
        "image",
        "audio",
        "video"
      ],
      "note": "MiMo-V2.5 is a native omnimodal model by Xiaomi.",
      "release_date": "2026-04",
      "popularity": 19
    },
    {
      "id": "mimo-v2-5-pro",
      "name": "MiMo-V2.5-Pro",
      "provider": "Xiaomi",
      "params": 1023.2,
      "active_params": 42,
      "is_moe": true,
      "kv_head_fraction": 0.0625,
      "arch": {
        "layers": 70,
        "hidden": 6144,
        "intermediate": 16384,
        "heads": 128,
        "kv_heads": 8,
        "head_dim": 192,
        "vocab": 152576,
        "tie_embeddings": false,
        "max_position": 1048576,
        "experts": 384,
        "experts_per_tok": 8,
        "moe_intermediate": 2048,
        "sliding_window": 128,
        "sliding_layers": 60
      },
      "ollama_tag": null,
      "hf_repo": "XiaomiMiMo/MiMo-V2.5-Pro",
      "modalities": [
        "text"
      ],
      "note": "MiMo-V2.5-Pro is Xiaomi’s flagship model, delivering strong performance in general agentic capabilities, complex software engineering, and long-horizon tasks, with top rankings on benchmarks such…",
      "release_date": "2026-04",
      "popularity": 5
    },
    {
      "id": "mimo-v2-6-flash",
      "name": "MiMo-V2.6-Flash",
      "provider": "Xiaomi",
      "params": 310.8,
      "active_params": 15,
      "is_moe": true,
      "kv_head_fraction": 0.0625,
      "arch": {
        "layers": 48,
        "hidden": 4096,
        "intermediate": 16384,
        "heads": 64,
        "kv_heads": 4,
        "head_dim": 192,
        "vocab": 152576,
        "tie_embeddings": false,
        "max_position": 1048576,
        "experts": 256,
        "experts_per_tok": 8,
        "moe_intermediate": 2048,
        "sliding_window": 128,
        "sliding_layers": 39
      },
      "ollama_tag": null,
      "hf_repo": "XiaomiMiMo/MiMo-V2.6-Flash-RL",
      "modalities": [
        "text",
        "image",
        "audio",
        "video"
      ],
      "note": "MiMo-V2.6-Flash is an open-source foundation model developed by Xiaomi.",
      "release_date": "2026-09",
      "popularity": 14
    }
  ]
}
