{
 "license": "CC BY 4.0",
 "attribution": "RigCheck, https://localrigcheck.com",
 "models": [
  {
   "url": "/deepseek-v4-1-flash/",
   "name": "DeepSeek V4.1 Flash",
   "updated": "2026-10-21",
   "checked": "2026-10-20",
   "subpages": [],
   "id": "deepseek-v4-1-flash",
   "role": "model",
   "parent": null,
   "checkers": [
    "deepseek"
   ],
   "modality": {
    "primary": "text",
    "inputs": [
     "text",
     "image"
    ],
    "outputs": [
     "text"
    ]
   },
   "family": "deepseek",
   "variants": [
    {
     "id": "deepseek-v4-1-flash",
     "label": "DeepSeek V4.1 Flash September release",
     "params": {
      "total": 552,
      "active": 16,
      "scope": "September 10, 2026; same config.json and size as 1017"
     }
    },
    {
     "id": "deepseek-v4-1-flash-1017",
     "label": "DeepSeek V4.1 Flash 1017 update",
     "params": {
      "total": 552,
      "active": 16,
      "scope": "October 17, 2026; backbone parameters (8B active in prefill, 16B in decode); with 196B of Engram tables, about 14B of MTP/DSpark weights and the vision encoder, 763.2B in the checkpoint"
     }
    }
   ],
   "dates": {
    "added": "2026-10-18",
    "announced": null,
    "weights_released": "2026-09-10",
    "weights_eta": null
   },
   "weights": "released",
   "routes": [
    {
     "runtime": "DwarfStar",
     "variants": "all",
     "where": "local",
     "status": "community",
     "requires": "its own Q2 GGUF (365.7 GB) on a fast SSD; SSD streaming on one 128 GB Mac or DGX Spark, documented for the September file (its 1017 file is not yet in its guides); its Q4 is September only",
     "evidence": "https://github.com/antirez/ds4/blob/main/docs/MODELS.md",
     "read": "2026-10-20"
    },
    {
     "runtime": "llama.cpp",
     "variants": "all",
     "where": "local",
     "status": "experimental",
     "requires": "an unmerged branch: Unsloth's pull request #30817 for its 1017 GGUF from 233.0 GB (Unsloth: 128 GB RAM setups such as a MacBook, DGX Spark or RTX Spark), or pull request #28696 for mrfakename's 1017 GGUF from 264.5 GB (no hardware named) or your own conversion of the September weights (Q2_K reported on one DGX Spark)",
     "evidence": "https://unsloth.ai/docs/models/deepseek-v4.1-flash",
     "read": "2026-10-20"
    },
    {
     "runtime": "oMLX",
     "variants": [
      "deepseek-v4-1-flash-1017"
     ],
     "where": "local",
     "status": "community",
     "requires": "Apple Silicon Mac; mlx-community oQ builds from 313.3 GB",
     "evidence": "https://huggingface.co/mlx-community/DeepSeek-V4.1-Flash-1017-oQ2e-mtp",
     "read": "2026-10-18"
    },
    {
     "runtime": "vLLM / SGLang",
     "variants": [
      "deepseek-v4-1-flash"
     ],
     "where": "local",
     "status": "community",
     "requires": "data-center GPUs: NVIDIA's NVFP4 checkpoint on four GB300; the current 429.0 GB revision needs vLLM pull request #61530, SGLang and stock vLLM load only the previous revision (about 492 GiB)",
     "evidence": "https://huggingface.co/nvidia/DeepSeek-V4.1-Flash-NVFP4",
     "read": "2026-10-18"
    },
    {
     "runtime": "Ollama",
     "variants": "all",
     "where": "cloud",
     "status": "cloud-only",
     "requires": null,
     "evidence": "https://ollama.com/library/deepseek-v4.1-flash/tags",
     "read": "2026-10-20"
    }
   ],
   "license": {
    "name": "MIT",
    "flags": []
   },
   "sheet": [
    {
     "key": "Developer",
     "value": "DeepSeek",
     "ev": "official",
     "src": 1
    },
    {
     "key": "Params",
     "value": "552B backbone; 8B active per token in prefill, 16B in decode; Engram conditional memory 196B",
     "ev": "official",
     "src": 1,
     "variant": "DeepSeek V4.1 Flash September release"
    },
    {
     "key": "Context",
     "value": "1M tokens (1,048,576)",
     "ev": "official",
     "src": 1
    },
    {
     "key": "Weights",
     "value": "September 10 release and the October 17 update, 510.3 GB each, not gated",
     "ev": "official",
     "src": 1
    },
    {
     "key": "License",
     "value": "MIT",
     "ev": "official",
     "src": 1
    },
    {
     "key": "llama.cpp",
     "value": "Not in mainline",
     "ev": "community",
     "src": 3
    },
    {
     "key": "DwarfStar",
     "value": "streams its 365.7 GB Q2 file of the September weights from SSD on one 128 GB Mac or DGX Spark",
     "ev": "community",
     "src": 4,
     "variant": "DeepSeek V4.1 Flash September release"
    },
    {
     "key": "Ollama",
     "value": "Ollama and LM Studio offer the model only in their clouds",
     "ev": "official",
     "src": 7
    }
   ]
  },
  {
   "url": "/flux-3-image-klein/",
   "name": "FLUX 3 Image Klein",
   "updated": "2026-10-19",
   "checked": "2026-10-17",
   "subpages": [],
   "id": "flux-3-image-klein",
   "role": "model",
   "parent": null,
   "checkers": [
    "flux"
   ],
   "modality": {
    "primary": "image",
    "inputs": [
     "text",
     "image"
    ],
    "outputs": [
     "image"
    ]
   },
   "family": "flux",
   "variants": [
    {
     "id": "flux-3-image-klein",
     "label": "FLUX 3 Image Klein",
     "params": {
      "total": 5.6,
      "scope": "image generator (text encoder not counted); distilled 16-step and base 50-step versions"
     }
    }
   ],
   "dates": {
    "announced": null,
    "weights_released": "2026-10-15",
    "weights_eta": null,
    "added": "2026-10-17"
   },
   "weights": "released",
   "routes": [
    {
     "runtime": "ComfyUI",
     "variants": "all",
     "where": "local",
     "status": "official",
     "requires": "BFL's int8 files on 8 GB+ NVIDIA GPUs; NVFP4 files also published (BFL's own NVFP4 code needs Blackwell)",
     "evidence": "https://huggingface.co/black-forest-labs/flux-3-image-klein-comfy",
     "read": "2026-10-17"
    },
    {
     "runtime": "BFL reference code",
     "variants": "all",
     "where": "local",
     "status": "official",
     "requires": "CUDA; BF16, FP8 (Ada, Hopper, Blackwell) or NVFP4 (Blackwell), no int8 path",
     "evidence": "https://github.com/black-forest-labs/flux-3-image-klein",
     "read": "2026-10-17"
    },
    {
     "runtime": "Diffusers",
     "variants": "all",
     "where": "local",
     "status": "official",
     "requires": "Modular Diffusers pipeline on Diffusers' main branch",
     "evidence": "https://github.com/huggingface/diffusers/tree/main/src/diffusers/modular_pipelines/flux3",
     "read": "2026-10-16"
    }
   ],
   "license": {
    "name": "FLUX Kommunity License",
    "flags": [
     "commercial-terms",
     "gated"
    ]
   },
   "sheet": [
    {
     "key": "Developer",
     "value": "Black Forest Labs",
     "ev": "official",
     "src": 1
    },
    {
     "key": "Params",
     "value": "5.6B (image generator; text encoder not counted)",
     "ev": "official",
     "src": 1
    },
    {
     "key": "Tasks",
     "value": "Text-to-image, editing with up to 10 reference images, layout with bounding boxes, up to 4K",
     "ev": "official",
     "src": 1
    },
    {
     "key": "License",
     "value": "FLUX Kommunity License",
     "ev": "official",
     "src": 1
    },
    {
     "key": "Weights",
     "value": "Hugging Face, gated (accept the license first)",
     "ev": "official",
     "src": 3
    },
    {
     "key": "ComfyUI",
     "value": "the int8 files in ComfyUI run on 8 GB GPUs such as the RTX 3070",
     "ev": "official",
     "src": 1
    },
    {
     "key": "BFL reference code",
     "value": "BFL's own code has no int8 path",
     "ev": "official",
     "src": 2
    },
    {
     "key": "Upsampler",
     "value": "BFL does not say whether its 8 GB figure includes the 9B upsampler",
     "ev": "official",
     "src": 2
    }
   ]
  },
  {
   "url": "/gemma-4/",
   "name": "Gemma 4",
   "updated": "2026-10-22",
   "checked": "2026-10-12",
   "subpages": [],
   "id": "gemma-4",
   "role": "model",
   "parent": null,
   "checkers": [
    "gemma4"
   ],
   "modality": {
    "primary": "text",
    "inputs": [
     "text",
     "image",
     "audio"
    ],
    "outputs": [
     "text"
    ]
   },
   "family": "gemma",
   "variants": [
    {
     "id": "gemma-4-e2b",
     "label": "Gemma 4 E2B",
     "params": {
      "total": 5.1,
      "scope": "2.3B effective, the rest per-layer embeddings"
     },
     "inputs": [
      "text",
      "image",
      "audio"
     ]
    },
    {
     "id": "gemma-4-e4b",
     "label": "Gemma 4 E4B",
     "params": {
      "total": 8,
      "scope": "4.5B effective, the rest per-layer embeddings"
     },
     "inputs": [
      "text",
      "image",
      "audio"
     ]
    },
    {
     "id": "gemma-4-12b",
     "label": "Gemma 4 12B",
     "params": {
      "total": 12,
      "scope": "dense"
     },
     "inputs": [
      "text",
      "image",
      "audio"
     ]
    },
    {
     "id": "gemma-4-26b-a4b",
     "label": "Gemma 4 26B-A4B",
     "params": {
      "total": 26,
      "active": 4,
      "scope": "mixture of experts"
     },
     "inputs": [
      "text",
      "image"
     ]
    },
    {
     "id": "gemma-4-31b",
     "label": "Gemma 4 31B",
     "params": {
      "total": 31,
      "scope": "dense"
     },
     "inputs": [
      "text",
      "image"
     ]
    }
   ],
   "dates": {
    "announced": null,
    "weights_released": null,
    "weights_eta": null,
    "added": "2026-10-09"
   },
   "weights": "released",
   "routes": [
    {
     "runtime": "llama.cpp",
     "variants": "all",
     "where": "local",
     "status": "community",
     "evidence": "https://unsloth.ai/docs/models/gemma-4",
     "read": "2026-10-12"
    },
    {
     "runtime": "Ollama",
     "variants": "all",
     "where": "local",
     "status": "official",
     "evidence": "https://ollama.com/library/gemma4/tags",
     "read": "2026-10-12"
    },
    {
     "runtime": "llama.cpp",
     "variants": [
      "gemma-4-e2b"
     ],
     "where": "local",
     "status": "official",
     "requires": "Jetson Orin Nano, NVIDIA's Jetson container",
     "evidence": "https://www.jetson-ai-lab.com/tutorials/gemma4-on-jetson/",
     "read": "2026-10-19"
    }
   ],
   "license": {
    "name": "Apache 2.0",
    "flags": []
   },
   "sheet": [
    {
     "key": "Developer",
     "value": "Google",
     "ev": "official",
     "src": 1
    },
    {
     "key": "Params",
     "value": "the dense 12B",
     "ev": "official",
     "src": 1,
     "variant": "Gemma 4 12B"
    },
    {
     "key": "Params",
     "value": "26B parameters, about 4B active per token",
     "ev": "official",
     "src": 1,
     "variant": "Gemma 4 26B-A4B"
    },
    {
     "key": "Params",
     "value": "the dense 31B",
     "ev": "official",
     "src": 1,
     "variant": "Gemma 4 31B"
    },
    {
     "key": "Context",
     "value": "131,072 tokens for E2B and E4B, 262,144 for the others",
     "ev": "official",
     "src": 1
    },
    {
     "key": "Input",
     "value": "text, image, audio",
     "ev": "official",
     "src": 1,
     "variant": "Gemma 4 12B"
    },
    {
     "key": "License",
     "value": "Apache 2.0",
     "ev": "official",
     "src": 1
    },
    {
     "key": "Ollama",
     "value": "Ollama has gemma4 packages",
     "ev": "official",
     "src": 3
    },
    {
     "key": "llama.cpp",
     "value": "llama.cpp supports the architecture, Unsloth publishes GGUF files",
     "ev": "community",
     "src": 2,
     "label": "Unsloth"
    }
   ]
  },
  {
   "url": "/glm-5-3-flash/",
   "name": "GLM-5.3-Flash",
   "updated": "2026-10-19",
   "checked": "2026-10-17",
   "subpages": [],
   "id": "glm-5-3-flash",
   "role": "model",
   "parent": null,
   "checkers": [
    "glm"
   ],
   "modality": {
    "primary": "text",
    "inputs": [
     "text",
     "image",
     "video"
    ],
    "outputs": [
     "text"
    ]
   },
   "family": "glm",
   "variants": [
    {
     "id": "glm-5-3-flash",
     "label": "GLM-5.3-Flash",
     "params": {
      "total": 320,
      "active": 18,
      "scope": "mixture of experts, 288 routed + 1 shared expert, 8 per token"
     }
    }
   ],
   "dates": {
    "announced": null,
    "weights_released": null,
    "weights_eta": null,
    "added": "2026-10-17"
   },
   "weights": "released",
   "routes": [
    {
     "runtime": "llama.cpp",
     "variants": "all",
     "where": "local",
     "status": "community",
     "requires": "Unsloth's GGUF files and its llama.cpp branch; about 100 GB of RAM + VRAM for 1-bit",
     "evidence": "https://unsloth.ai/docs/models/glm-5.3-flash",
     "read": "2026-10-17"
    },
    {
     "runtime": "KTransformers",
     "variants": "all",
     "where": "local",
     "status": "community",
     "requires": "official FP8 weights, an RTX 40 or 50 card, an AVX-512 CPU, at least 350 GB of system memory",
     "evidence": "https://github.com/kvcache-ai/ktransformers/blob/main/doc/en/kt-kernel/GLM-5.3-Flash-Tutorial.md",
     "read": "2026-10-17"
    },
    {
     "runtime": "vLLM",
     "variants": "all",
     "where": "local",
     "status": "official",
     "requires": "server GPUs: NVIDIA Hopper or newer, AMD Instinct MI350X / MI355X",
     "evidence": "https://recipes.vllm.ai/zai-org/GLM-5.3-Flash",
     "read": "2026-10-09"
    },
    {
     "runtime": "SGLang",
     "variants": "all",
     "where": "local",
     "status": "official",
     "requires": "server GPUs (H100 to GB300, MI355X)",
     "evidence": "https://cookbook.sglang.io/autoregressive/GLM/GLM-5.3-Flash",
     "read": "2026-10-09"
    },
    {
     "runtime": "Ollama",
     "variants": "all",
     "where": "cloud",
     "status": "cloud-only",
     "requires": "glm-5.3-flash:cloud runs on Ollama's servers",
     "evidence": "https://ollama.com/library/glm-5.3-flash/tags",
     "read": "2026-10-17"
    }
   ],
   "license": {
    "name": "MIT",
    "flags": []
   },
   "sheet": [
    {
     "key": "Developer",
     "value": "Z.ai",
     "ev": "official",
     "src": 1
    },
    {
     "key": "Params",
     "value": "320B total, 18B active per token",
     "ev": "official",
     "src": 1
    },
    {
     "key": "Context",
     "value": "1,048,576 tokens",
     "ev": "official",
     "src": 1
    },
    {
     "key": "Official weights",
     "value": "FP8, 328.3 GB",
     "ev": "official",
     "src": 1
    },
    {
     "key": "License",
     "value": "MIT",
     "ev": "official",
     "src": 1
    },
    {
     "key": "llama.cpp",
     "value": "with llama.cpp and Unsloth's GGUF files",
     "ev": "community",
     "src": 2
    },
    {
     "key": "KTransformers",
     "value": "KTransformers runs them with one RTX 40 or 50 card (its tutorial gives no VRAM minimum), at least 350 GB of free system memory and a CPU with AVX-512",
     "ev": "official",
     "src": 3
    },
    {
     "key": "vLLM / SGLang",
     "value": "vLLM and SGLang target server GPUs (H100 and newer, AMD Instinct MI350X/MI355X)",
     "ev": "official",
     "src": 3
    },
    {
     "key": "Ollama",
     "value": "Only as glm-5.3-flash:cloud, which runs on Ollama's servers, not on your hardware",
     "ev": "official",
     "src": 4
    }
   ]
  },
  {
   "url": "/kolibri-1/",
   "name": "Kolibri-1",
   "updated": "2026-10-21",
   "checked": "2026-10-12",
   "subpages": [],
   "id": "kolibri-1",
   "role": "model",
   "parent": null,
   "checkers": [
    "kolibri"
   ],
   "modality": {
    "primary": "text",
    "inputs": [
     "text"
    ],
    "outputs": [
     "text"
    ]
   },
   "family": "aleph-alpha",
   "variants": [
    {
     "id": "kolibri-1",
     "label": "Kolibri-1",
     "params": {
      "total": 78.1,
      "active": 3.46,
      "scope": "mixture of experts, 384 routed + 1 shared expert, 6 selected per token"
     }
    }
   ],
   "dates": {
    "announced": null,
    "weights_released": null,
    "weights_eta": null,
    "added": "2026-10-12"
   },
   "weights": "released",
   "routes": [
    {
     "runtime": "vLLM",
     "variants": "all",
     "where": "local",
     "status": "official",
     "requires": "needs the aleph-alpha-inference plugin, FP8, data-center GPUs",
     "evidence": "https://github.com/Aleph-Alpha/aleph-alpha-inference",
     "read": "2026-10-12"
    },
    {
     "runtime": "Transformers",
     "variants": "all",
     "where": "local",
     "status": "official",
     "requires": "transformers 5.20 or newer; the official card gives no separate hardware for this route",
     "evidence": "https://huggingface.co/Aleph-Alpha/Kolibri-1",
     "read": "2026-10-21"
    },
    {
     "runtime": "llama.cpp",
     "variants": "all",
     "where": "local",
     "status": "patched",
     "requires": "independent Hob-forge patch, Q4_K_M GGUF",
     "evidence": "https://huggingface.co/Hob-forge/Kolibri-1-GGUF",
     "read": "2026-10-12"
    }
   ],
   "license": {
    "name": "Apache 2.0",
    "flags": []
   },
   "sheet": [
    {
     "key": "Developer",
     "value": "Aleph Alpha",
     "ev": "official",
     "src": 1
    },
    {
     "key": "Params",
     "value": "78.10B total parameters",
     "ev": "official",
     "src": 1
    },
    {
     "key": "Active",
     "value": "3.46B active per token",
     "ev": "official",
     "src": 1
    },
    {
     "key": "Weights",
     "value": "32 downloadable shards total 78.84 GB",
     "ev": "official",
     "src": 1
    },
    {
     "key": "License",
     "value": "Apache 2.0",
     "ev": "official",
     "src": 1
    },
    {
     "key": "Official routes",
     "value": "Aleph Alpha's vLLM plugin and, since October 20, 2026, Transformers 5.20 or newer",
     "ev": "official",
     "src": 1
    },
    {
     "key": "llama.cpp",
     "value": "requires a custom llama.cpp patch",
     "ev": "community",
     "src": 2
    },
    {
     "key": "Official minimum",
     "value": "2× A100 80 GB, 2× H100 SXM5, or 1× H200/B200/B300",
     "ev": "official",
     "src": 1
    }
   ]
  },
  {
   "url": "/lfm-2-5-24b-a2b/",
   "name": "LFM2.5-24B-A2B",
   "updated": "2026-10-22",
   "checked": "2026-10-22",
   "subpages": [],
   "id": "lfm-2-5-24b-a2b",
   "role": "model",
   "parent": null,
   "checkers": [],
   "modality": {
    "primary": "text",
    "inputs": [
     "text"
    ],
    "outputs": [
     "text"
    ]
   },
   "family": "lfm",
   "variants": [
    {
     "id": "lfm-2-5-24b-a2b",
     "label": "LFM2.5-24B-A2B",
     "params": {
      "total": 24.0,
      "active": 2.45,
      "scope": "mixture of experts, text only"
     }
    }
   ],
   "dates": {
    "announced": "2026-10-13",
    "weights_released": "2026-10-13",
    "weights_eta": null,
    "added": "2026-10-22"
   },
   "weights": "released",
   "routes": [
    {
     "runtime": "llama.cpp",
     "variants": "all",
     "where": "local",
     "status": "official",
     "evidence": "https://huggingface.co/LiquidAI/LFM2.5-24B-A2B-GGUF",
     "read": "2026-10-22"
    },
    {
     "runtime": "Ollama",
     "variants": "all",
     "where": "local",
     "status": "official",
     "evidence": "https://ollama.com/library/lfm2.5/tags",
     "read": "2026-10-22"
    },
    {
     "runtime": "MLX",
     "variants": "all",
     "where": "local",
     "status": "official",
     "evidence": "https://huggingface.co/LiquidAI/LFM2.5-24B-A2B",
     "read": "2026-10-22"
    },
    {
     "runtime": "LM Studio",
     "variants": "all",
     "where": "local",
     "status": "official",
     "evidence": "https://huggingface.co/LiquidAI/LFM2.5-24B-A2B",
     "read": "2026-10-22"
    },
    {
     "runtime": "vLLM",
     "variants": "all",
     "where": "local",
     "status": "official",
     "evidence": "https://huggingface.co/LiquidAI/LFM2.5-24B-A2B",
     "read": "2026-10-22"
    },
    {
     "runtime": "SGLang",
     "variants": "all",
     "where": "local",
     "status": "official",
     "evidence": "https://huggingface.co/LiquidAI/LFM2.5-24B-A2B",
     "read": "2026-10-22"
    },
    {
     "runtime": "Transformers",
     "variants": "all",
     "where": "local",
     "status": "official",
     "evidence": "https://huggingface.co/LiquidAI/LFM2.5-24B-A2B",
     "read": "2026-10-22"
    }
   ],
   "license": {
    "name": "LFM Open License v1.0",
    "flags": [
     "commercial-terms"
    ]
   },
   "sheet": [
    {
     "key": "Developer",
     "value": "Liquid AI",
     "ev": "official",
     "src": 1
    },
    {
     "key": "Params",
     "value": "24.0B total, 2.45B active per token",
     "ev": "official",
     "src": 1
    },
    {
     "key": "Context",
     "value": "131,072 tokens",
     "ev": "official",
     "src": 1
    },
    {
     "key": "Input",
     "value": "text only",
     "ev": "official",
     "src": 1
    },
    {
     "key": "License",
     "value": "LFM Open License v1.0",
     "ev": "official",
     "src": 1
    },
    {
     "key": "KV cache",
     "value": "20 KiB per token at f16",
     "ev": "estimate",
     "src": 1,
     "label": "RigCheck calculation"
    },
    {
     "key": "llama.cpp",
     "value": "Liquid AI publishes GGUF files for llama.cpp",
     "ev": "official",
     "src": 2
    },
    {
     "key": "Ollama",
     "value": "Ollama has lfm2.5:24b",
     "ev": "official",
     "src": 4
    }
   ]
  },
  {
   "url": "/minimax-h3/",
   "name": "MiniMax H3",
   "updated": "2026-10-22",
   "checked": "2026-10-12",
   "subpages": [],
   "id": "minimax-h3",
   "role": "model",
   "parent": null,
   "checkers": [
    "h3"
   ],
   "modality": {
    "primary": "video",
    "inputs": [
     "text",
     "image",
     "video",
     "audio"
    ],
    "outputs": [
     "video",
     "audio"
    ]
   },
   "family": "minimax",
   "variants": [
    {
     "id": "minimax-h3",
     "label": "MiniMax H3",
     "params": {
      "total": 33,
      "scope": "dense transformer only; the Qwen3-VL-32B text encoder is separate"
     }
    }
   ],
   "dates": {
    "announced": null,
    "weights_released": null,
    "weights_eta": null,
    "added": "2026-10-12"
   },
   "weights": "released",
   "routes": [
    {
     "runtime": "SGLang",
     "variants": "all",
     "where": "local",
     "status": "official",
     "evidence": "https://docs.sglang.io/cookbook/diffusion/MiniMax/MiniMax-H3",
     "read": "2026-10-12"
    },
    {
     "runtime": "vLLM-Omni",
     "variants": "all",
     "where": "local",
     "status": "official",
     "evidence": "https://recipes.vllm.ai/MiniMaxAI/MiniMax-H3",
     "read": "2026-10-12"
    },
    {
     "runtime": "Diffusers",
     "variants": "all",
     "where": "local",
     "status": "official",
     "evidence": "https://huggingface.co/docs/diffusers/main/en/api/pipelines/minimax_h3",
     "read": "2026-10-12"
    },
    {
     "runtime": "ComfyUI",
     "variants": "all",
     "where": "local",
     "status": "community",
     "evidence": "https://docs.comfy.org/tutorials/video/minimax/minimax-h3",
     "read": "2026-10-19"
    },
    {
     "runtime": "stable-diffusion.cpp",
     "variants": "all",
     "where": "local",
     "status": "community",
     "evidence": "https://huggingface.co/unsloth/MiniMax-H3-GGUF",
     "read": "2026-10-12"
    }
   ],
   "license": {
    "name": "MiniMax H3 Community License",
    "flags": [
     "region-restricted",
     "commercial-terms"
    ]
   },
   "sheet": [
    {
     "key": "Developer",
     "value": "MiniMax",
     "ev": "official",
     "src": 1
    },
    {
     "key": "Params",
     "value": "33B-parameter dense",
     "ev": "official",
     "src": 1
    },
    {
     "key": "Output",
     "value": "768-pixel short edge (1344×768 at 16:9), 24 fps, 4–15 seconds",
     "ev": "official",
     "src": 1
    },
    {
     "key": "License",
     "value": "MiniMax's open-weights license does not cover the US, the EU, the UK or South Korea",
     "ev": "official",
     "src": 1
    },
    {
     "key": "SGLang",
     "value": "12 GB of VRAM and 32 GB of RAM plus an NVMe drive at 480P (SGLang, run with the memory capped; a real 32 GB PC will be slower)",
     "ev": "official",
     "src": 2
    },
    {
     "key": "Diffusers",
     "value": "two 48 GB cards with int8 (diffusers)",
     "ev": "official",
     "src": 3
    },
    {
     "key": "stable-diffusion.cpp",
     "value": "Unsloth's GGUF files run in stable-diffusion.cpp",
     "ev": "community",
     "src": 5
    }
   ]
  },
  {
   "url": "/mistral-large-4/",
   "name": "Mistral Large 4",
   "updated": "2026-10-19",
   "checked": "2026-10-08",
   "subpages": [],
   "id": "mistral-large-4",
   "role": "model",
   "parent": null,
   "checkers": [
    "large4"
   ],
   "modality": {
    "primary": "text",
    "inputs": [
     "text",
     "image"
    ],
    "outputs": [
     "text"
    ]
   },
   "family": "mistral",
   "variants": [
    {
     "id": "mistral-large-4",
     "label": "Mistral Large 4",
     "params": {
      "total": 1050,
      "active": 52,
      "scope": "49B active without embeddings and output layers; a 1.6B vision encoder is listed separately"
     }
    }
   ],
   "dates": {
    "announced": "2026-10-06",
    "weights_released": null,
    "weights_eta": {
     "date": "2026-10-31",
     "source": "https://huggingface.co/mistralai/Mistral-Large-4.0-1T05-A52B",
     "read": "2026-10-08"
    },
    "added": "2026-10-07"
   },
   "weights": "announced",
   "routes": [
    {
     "runtime": "Mistral API",
     "variants": "all",
     "where": "cloud",
     "status": "official",
     "evidence": "https://docs.mistral.ai/models/mistral-large-4",
     "read": "2026-10-08"
    },
    {
     "runtime": "Ollama",
     "variants": "all",
     "where": "cloud",
     "status": "cloud-only",
     "evidence": "https://ollama.com/library/mistral-large-4/tags",
     "read": "2026-10-08"
    }
   ],
   "license": {
    "name": null,
    "flags": []
   },
   "sheet": [
    {
     "key": "Developer",
     "value": "Mistral AI",
     "ev": "official",
     "src": 2
    },
    {
     "key": "Total parameters",
     "value": "1.05 trillion (Mistral docs)",
     "ev": "official",
     "src": 1
    },
    {
     "key": "Active parameters",
     "value": "52B per token",
     "ev": "official",
     "src": 1
    },
    {
     "key": "Context",
     "value": "1M tokens (Mistral docs, API model page)",
     "ev": "official",
     "src": 1
    },
    {
     "key": "Weights",
     "value": "open weights announced for the end of October 2026",
     "ev": "official",
     "src": 2
    },
    {
     "key": "API",
     "value": "mistral-large-4-0",
     "ev": "official",
     "src": 1
    },
    {
     "key": "Ollama",
     "value": "only the cloud tag mistral-large-4:cloud",
     "ev": "official",
     "src": 4
    }
   ]
  },
  {
   "url": "/mistral-medium-3-5/",
   "name": "Mistral Medium 3.5",
   "updated": "2026-10-19",
   "checked": "2026-10-12",
   "subpages": [],
   "id": "mistral-medium-3-5",
   "role": "model",
   "parent": null,
   "checkers": [
    "medium"
   ],
   "modality": {
    "primary": "text",
    "inputs": [
     "text",
     "image"
    ],
    "outputs": [
     "text"
    ]
   },
   "family": "mistral",
   "variants": [
    {
     "id": "mistral-medium-3-5",
     "label": "Mistral Medium 3.5",
     "params": {
      "total": 128,
      "scope": "dense; a vision encoder handles image input"
     }
    }
   ],
   "dates": {
    "announced": null,
    "weights_released": "2026-04-28",
    "weights_eta": null,
    "added": "2026-10-08"
   },
   "weights": "released",
   "routes": [
    {
     "runtime": "llama.cpp",
     "variants": "all",
     "where": "local",
     "status": "community",
     "evidence": "https://unsloth.ai/docs/models/mistral-3.5",
     "read": "2026-10-12"
    },
    {
     "runtime": "Ollama",
     "variants": "all",
     "where": "local",
     "status": "official",
     "evidence": "https://ollama.com/library/mistral-medium-3.5/tags",
     "read": "2026-10-12"
    },
    {
     "runtime": "vLLM",
     "variants": "all",
     "where": "local",
     "status": "official",
     "evidence": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B",
     "read": "2026-10-12"
    },
    {
     "runtime": "SGLang",
     "variants": "all",
     "where": "local",
     "status": "official",
     "evidence": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B",
     "read": "2026-10-12"
    },
    {
     "runtime": "Transformers",
     "variants": "all",
     "where": "local",
     "status": "official",
     "evidence": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B",
     "read": "2026-10-12"
    }
   ],
   "license": {
    "name": "Modified MIT",
    "flags": [
     "commercial-terms"
    ]
   },
   "sheet": [
    {
     "key": "Developer",
     "value": "Mistral AI",
     "ev": "official",
     "src": 1
    },
    {
     "key": "Params",
     "value": "128B, dense",
     "ev": "official",
     "src": 1
    },
    {
     "key": "Context",
     "value": "256K (262,144 tokens)",
     "ev": "official",
     "src": 1
    },
    {
     "key": "Official weights",
     "value": "FP8 (e4m3), 133.6 GB",
     "ev": "official",
     "src": 1
    },
    {
     "key": "KV cache",
     "value": "352 KiB per token at f16",
     "ev": "estimate",
     "src": 1,
     "label": "RigCheck calculation"
    },
    {
     "key": "License",
     "value": "Modified MIT",
     "ev": "official",
     "src": 1
    },
    {
     "key": "Release",
     "value": "April 28, 2026, as open weights",
     "ev": "official",
     "src": 5
    },
    {
     "key": "vLLM",
     "value": "vLLM (recommended by Mistral)",
     "ev": "official",
     "src": 1
    }
   ]
  },
  {
   "url": "/nemotron-3-5/",
   "name": "Nemotron 3.5",
   "updated": "2026-10-20",
   "checked": "2026-10-17",
   "subpages": [],
   "id": "nemotron-3-5",
   "role": "model",
   "parent": null,
   "checkers": [
    "nemotron"
   ],
   "modality": {
    "primary": "text",
    "inputs": [
     "text",
     "image",
     "video"
    ],
    "outputs": [
     "text"
    ]
   },
   "family": "nemotron",
   "variants": [
    {
     "id": "nemotron-3-5-bolt",
     "label": "Nemotron 3.5 Bolt 74B-A9B",
     "params": {
      "total": 74,
      "active": 9,
      "scope": "mixture of experts, hybrid Mamba-2 / transformer"
     }
    },
    {
     "id": "nemotron-3-5-super-vl",
     "label": "Nemotron 3.5 Super VL 120B-A12B",
     "params": {
      "total": 120,
      "active": 12,
      "scope": "mixture of experts, hybrid Mamba-2 / transformer"
     }
    }
   ],
   "dates": {
    "announced": null,
    "weights_released": "2026-10-15",
    "weights_eta": null,
    "added": "2026-10-17"
   },
   "weights": "released",
   "routes": [
    {
     "runtime": "llama.cpp",
     "variants": [
      "nemotron-3-5-bolt"
     ],
     "where": "local",
     "status": "official",
     "requires": "NVIDIA's 2-bit GGUF and recipes: a 32 GB+ RTX 40/50 GPU or an RTX Spark; latest llama.cpp built from source",
     "evidence": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB-GGUF",
     "read": "2026-10-17"
    },
    {
     "runtime": "llama.cpp",
     "variants": [
      "nemotron-3-5-super-vl"
     ],
     "where": "local",
     "status": "community",
     "requires": "ggml-org and Unsloth GGUF files, 64 GB+ of RAM + VRAM or unified memory",
     "evidence": "https://unsloth.ai/docs/models/nemotron-3.5-super-vl",
     "read": "2026-10-17"
    },
    {
     "runtime": "vLLM",
     "variants": [
      "nemotron-3-5-bolt"
     ],
     "where": "local",
     "status": "official",
     "requires": "DGX Spark: the 24GB packed build, 256K context, fp8 KV cache, a nightly vLLM image at release",
     "evidence": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB",
     "read": "2026-10-17"
    },
    {
     "runtime": "vLLM",
     "variants": [
      "nemotron-3-5-super-vl"
     ],
     "where": "local",
     "status": "official",
     "requires": "NVFP4 weights on one DGX Spark, one H200, two H100s or one GB200",
     "evidence": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-NVFP4",
     "read": "2026-10-17"
    },
    {
     "runtime": "Ollama",
     "variants": [
      "nemotron-3-5-bolt"
     ],
     "where": "local",
     "status": "official",
     "requires": "nemotron-3.5-bolt, 25 GB",
     "evidence": "https://ollama.com/library/nemotron-3.5-bolt/tags",
     "read": "2026-10-17"
    },
    {
     "runtime": "Ollama",
     "variants": [
      "nemotron-3-5-super-vl"
     ],
     "where": "local",
     "status": "official",
     "requires": "nemotron-3.5-super, 80–249 GB",
     "evidence": "https://ollama.com/library/nemotron-3.5-super/tags",
     "read": "2026-10-17"
    },
    {
     "runtime": "LM Studio",
     "variants": [
      "nemotron-3-5-super-vl"
     ],
     "where": "local",
     "status": "official",
     "requires": "at least 70 GB of system memory (LM Studio's listing)",
     "evidence": "https://lmstudio.ai/models/nvidia/nemotron-3.5-super-vl",
     "read": "2026-10-17"
    }
   ],
   "license": {
    "name": "OpenMDW-1.1",
    "flags": []
   },
   "sheet": [
    {
     "key": "Developer",
     "value": "NVIDIA",
     "ev": "official",
     "src": 1
    },
    {
     "key": "Params",
     "value": "74B total, 9B active per token",
     "ev": "official",
     "src": 1,
     "variant": "Nemotron 3.5 Bolt 74B-A9B"
    },
    {
     "key": "Params",
     "value": "120B total, 12B active",
     "ev": "official",
     "src": 2,
     "variant": "Nemotron 3.5 Super VL 120B-A12B"
    },
    {
     "key": "Context",
     "value": "up to 256K tokens (262,144)",
     "ev": "official",
     "src": 1,
     "variant": "Nemotron 3.5 Bolt 74B-A9B"
    },
    {
     "key": "Context",
     "value": "up to 1M tokens (1,048,576)",
     "ev": "official",
     "src": 2,
     "variant": "Nemotron 3.5 Super VL 120B-A12B"
    },
    {
     "key": "Release",
     "value": "October 15, 2026 on Hugging Face (nvidia organization, not gated)",
     "ev": "official",
     "src": 1
    },
    {
     "key": "License",
     "value": "OpenMDW-1.1",
     "ev": "official",
     "src": 1
    },
    {
     "key": "llama.cpp",
     "value": "NVIDIA's own llama.cpp recipes start at a 32 GB GPU (RTX 5090)",
     "ev": "official",
     "src": 1,
     "variant": "Nemotron 3.5 Bolt 74B-A9B"
    }
   ]
  },
  {
   "url": "/qwen-3-8-27b/",
   "name": "Qwen3.8-27B",
   "updated": "2026-10-22",
   "checked": "2026-10-12",
   "subpages": [],
   "id": "qwen-3-8-27b",
   "role": "model",
   "parent": null,
   "checkers": [
    "qwen27"
   ],
   "modality": {
    "primary": "text",
    "inputs": [
     "text",
     "image",
     "video"
    ],
    "outputs": [
     "text"
    ]
   },
   "family": "qwen",
   "variants": [
    {
     "id": "qwen-3-8-27b",
     "label": "Qwen3.8-27B",
     "params": {
      "total": 27,
      "scope": "dense language model; the vision encoder is part of the release"
     }
    }
   ],
   "dates": {
    "announced": null,
    "weights_released": null,
    "weights_eta": null,
    "added": "2026-10-09"
   },
   "weights": "released",
   "routes": [
    {
     "runtime": "llama.cpp",
     "variants": "all",
     "where": "local",
     "status": "community",
     "evidence": "https://unsloth.ai/docs/models/qwen3.8",
     "read": "2026-10-12"
    },
    {
     "runtime": "Ollama",
     "variants": "all",
     "where": "local",
     "status": "official",
     "evidence": "https://ollama.com/library/qwen3.8/tags",
     "read": "2026-10-12"
    },
    {
     "runtime": "vLLM",
     "variants": "all",
     "where": "local",
     "status": "official",
     "evidence": "https://huggingface.co/Qwen/Qwen3.8-27B",
     "read": "2026-10-12"
    },
    {
     "runtime": "SGLang",
     "variants": "all",
     "where": "local",
     "status": "official",
     "evidence": "https://huggingface.co/Qwen/Qwen3.8-27B",
     "read": "2026-10-12"
    },
    {
     "runtime": "Transformers",
     "variants": "all",
     "where": "local",
     "status": "official",
     "evidence": "https://huggingface.co/Qwen/Qwen3.8-27B",
     "read": "2026-10-12"
    }
   ],
   "license": {
    "name": "Apache 2.0",
    "flags": []
   },
   "sheet": [
    {
     "key": "Developer",
     "value": "Qwen",
     "ev": "official",
     "src": 1
    },
    {
     "key": "Params",
     "value": "27B, dense",
     "ev": "official",
     "src": 1
    },
    {
     "key": "Context",
     "value": "262,144 tokens natively",
     "ev": "official",
     "src": 1
    },
    {
     "key": "Input",
     "value": "vision encoder for image and video input",
     "ev": "official",
     "src": 1
    },
    {
     "key": "License",
     "value": "Apache 2.0",
     "ev": "official",
     "src": 1
    },
    {
     "key": "KV cache",
     "value": "64 KiB per token at f16",
     "ev": "estimate",
     "src": 1,
     "label": "RigCheck calculation"
    },
    {
     "key": "Ollama",
     "value": "Ollama has qwen3.8:27b",
     "ev": "official",
     "src": 3
    },
    {
     "key": "llama.cpp",
     "value": "llama.cpp supports the architecture, Unsloth publishes GGUF files",
     "ev": "community",
     "src": 2,
     "label": "Unsloth"
    }
   ]
  },
  {
   "url": "/qwen-3-8-flash-next/",
   "name": "Qwen3.8-Flash-Next",
   "updated": "2026-10-19",
   "checked": "2026-10-08",
   "subpages": [],
   "id": "qwen-3-8-flash-next",
   "role": "model",
   "parent": null,
   "checkers": [
    "qwen"
   ],
   "modality": {
    "primary": "text",
    "inputs": [
     "text",
     "image",
     "video"
    ],
    "outputs": [
     "text"
    ]
   },
   "family": "qwen",
   "variants": [
    {
     "id": "qwen-3-8-flash-next",
     "label": "Qwen3.8-Flash-Next",
     "params": {
      "total": 125,
      "active": 6,
      "scope": "language model; the 51B n-gram embedding table and 4B MTP layer are counted separately"
     }
    }
   ],
   "dates": {
    "announced": null,
    "weights_released": null,
    "weights_eta": null,
    "added": "2026-10-08"
   },
   "weights": "released",
   "routes": [
    {
     "runtime": "llama.cpp",
     "variants": "all",
     "where": "local",
     "status": "community",
     "evidence": "https://unsloth.ai/docs/models/qwen3.8-next",
     "read": "2026-10-08"
    },
    {
     "runtime": "Ollama",
     "variants": "all",
     "where": "local",
     "status": "official",
     "evidence": "https://ollama.com/library/qwen3.8-flash-next/tags",
     "read": "2026-10-08"
    },
    {
     "runtime": "LM Studio",
     "variants": "all",
     "where": "local",
     "status": "community",
     "evidence": "https://huggingface.co/lmstudio-community/Qwen3.8-Flash-Next-GGUF",
     "read": "2026-10-08"
    },
    {
     "runtime": "vLLM",
     "variants": "all",
     "where": "local",
     "status": "official",
     "evidence": "https://recipes.vllm.ai/Qwen/Qwen3.8-Flash-Next",
     "read": "2026-10-08"
    },
    {
     "runtime": "SGLang",
     "variants": "all",
     "where": "local",
     "status": "official",
     "evidence": "https://docs.sglang.io/cookbook/autoregressive/Qwen/Qwen3.8-Flash-Next",
     "read": "2026-10-08"
    },
    {
     "runtime": "Strata",
     "variants": "all",
     "where": "local",
     "status": "community",
     "evidence": "/strata/",
     "read": "2026-10-08"
    }
   ],
   "license": {
    "name": "Qwen Community License 1.0",
    "flags": [
     "commercial-terms"
    ]
   },
   "sheet": [
    {
     "key": "Developer",
     "value": "Qwen",
     "ev": "official",
     "src": 1
    },
    {
     "key": "Params",
     "value": "125B in the language model, 6B active per token; plus 51B n-gram embedding and 4B MTP",
     "ev": "official",
     "src": 1
    },
    {
     "key": "Context",
     "value": "262,144 tokens natively",
     "ev": "official",
     "src": 1
    },
    {
     "key": "Official weights",
     "value": "BF16, 360.0 GB; FP8, 185.5 GB",
     "ev": "official",
     "src": 1
    },
    {
     "key": "License",
     "value": "Qwen Community License 1.0",
     "ev": "official",
     "src": 1
    },
    {
     "key": "Ollama",
     "value": "Ollama packages its own builds of qwen3.8-flash-next",
     "ev": "official",
     "src": 4
    },
    {
     "key": "llama.cpp",
     "value": "llama.cpp supports the architecture",
     "ev": "official",
     "src": 6,
     "label": "llama.cpp"
    },
    {
     "key": "vLLM",
     "value": "Server GPUs, for the official weights",
     "ev": "official",
     "src": 5
    }
   ]
  },
  {
   "url": "/qwen-image-2-1/",
   "name": "Qwen-Image-2.1",
   "updated": "2026-10-19",
   "checked": "2026-10-12",
   "subpages": [],
   "id": "qwen-image-2-1",
   "role": "model",
   "parent": null,
   "checkers": [
    "image"
   ],
   "modality": {
    "primary": "image",
    "inputs": [
     "text",
     "image"
    ],
    "outputs": [
     "image"
    ]
   },
   "family": "qwen",
   "variants": [
    {
     "id": "qwen-image-2-1",
     "label": "Qwen-Image-2.1 (Base and Turbo)",
     "params": {
      "total": 7,
      "scope": "visual generation model only; the Qwen3-VL-8B encoder and the VAE are separate"
     }
    }
   ],
   "dates": {
    "announced": null,
    "weights_released": null,
    "weights_eta": null,
    "added": "2026-10-12"
   },
   "weights": "released",
   "routes": [
    {
     "runtime": "ComfyUI",
     "variants": "all",
     "where": "local",
     "status": "community",
     "requires": "Base; Turbo sampling not verified here",
     "evidence": "https://huggingface.co/Comfy-Org/Qwen-Image-2.1",
     "read": "2026-10-12"
    },
    {
     "runtime": "Diffusers",
     "variants": "all",
     "where": "local",
     "status": "official",
     "requires": "Turbo needs Diffusers from GitHub (PR #14950)",
     "evidence": "https://huggingface.co/Qwen/Qwen-Image-2.1-Turbo",
     "read": "2026-10-12"
    },
    {
     "runtime": "stable-diffusion.cpp",
     "variants": "all",
     "where": "local",
     "status": "community",
     "requires": "Base GGUF; Turbo GGUF route not verified here",
     "evidence": "https://github.com/leejet/stable-diffusion.cpp/blob/master/docs/qwen_image_2.1.md",
     "read": "2026-10-12"
    }
   ],
   "license": {
    "name": "Qwen Research License",
    "flags": [
     "research-only"
    ]
   },
   "sheet": [
    {
     "key": "Developer",
     "value": "Qwen",
     "ev": "official",
     "src": 1
    },
    {
     "key": "Params",
     "value": "7B visual generation architecture",
     "ev": "official",
     "src": 1
    },
    {
     "key": "Components",
     "value": "Qwen3-VL-8B text/vision encoder and a model-specific VAE",
     "ev": "official",
     "src": 1
    },
    {
     "key": "Files",
     "value": "ComfyUI BF16 files total 32.44 GB",
     "ev": "community",
     "src": 2
    },
    {
     "key": "License",
     "value": "Qwen Research License",
     "ev": "official",
     "src": 1
    },
    {
     "key": "ComfyUI",
     "value": "Base CUDA ComfyUI and official Diffusers routes are documented; Turbo ComfyUI sampling remains unverified here",
     "ev": "community",
     "src": 2
    },
    {
     "key": "stable-diffusion.cpp",
     "value": "Base GGUF has a stable-diffusion.cpp route",
     "ev": "community",
     "src": 3
    }
   ]
  },
  {
   "url": "/strata/",
   "name": "Strata checker",
   "updated": "2026-10-21",
   "checked": "2026-10-21",
   "subpages": [],
   "id": "strata",
   "role": "engine",
   "parent": null,
   "checkers": [
    "strata"
   ],
   "modality": {
    "primary": "text",
    "inputs": [
     "text",
     "image"
    ],
    "outputs": [
     "text"
    ]
   },
   "family": null,
   "variants": [],
   "dates": {
    "announced": null,
    "weights_released": null,
    "weights_eta": null,
    "added": "2026-10-07"
   },
   "weights": null,
   "runs": [
    "qwen-3-8-flash-next"
   ],
   "routes": [
    {
     "runtime": "Strata",
     "variants": "all",
     "where": "local",
     "status": "official",
     "requires": "Windows / Linux, NVIDIA RTX 20+ or listed AMD, 32 GB of RAM and VRAM together",
     "evidence": "https://github.com/Niko1221/Strata",
     "read": "2026-10-21"
    },
    {
     "runtime": "Strata",
     "variants": "all",
     "where": "local",
     "status": "experimental",
     "requires": "Apple Silicon Mac (Metal, ready-made in the Mac app), macOS 15 or newer",
     "evidence": "https://github.com/Niko1221/Strata/blob/main/docs/MAC.md",
     "read": "2026-10-21"
    },
    {
     "runtime": "Strata",
     "variants": "all",
     "where": "local",
     "status": "experimental",
     "requires": "older NVIDIA cards such as the GTX 10 series or Tesla P40 / V100 (separate CUDA 12 engine, Windows or Linux)",
     "evidence": "https://github.com/Niko1221/Strata/blob/main/docs/OLDER_GPUS.md",
     "read": "2026-10-21"
    },
    {
     "runtime": "Strata",
     "variants": "all",
     "where": "local",
     "status": "experimental",
     "requires": "older AMD cards such as the Radeon VII / MI50 (Linux build)",
     "evidence": "https://github.com/Niko1221/Strata/blob/main/docs/OLDER_GPUS.md",
     "read": "2026-10-21"
    },
    {
     "runtime": "Strata",
     "variants": "all",
     "where": "local",
     "status": "experimental",
     "requires": "Intel Arc (experimental Linux build)",
     "evidence": "https://github.com/Niko1221/Strata",
     "read": "2026-10-21"
    }
   ],
   "license": {
    "name": "MIT (engine)",
    "flags": []
   },
   "sheet": [
    {
     "key": "Type",
     "value": "Strata is an inference engine, not a model",
     "ev": "official",
     "src": 1
    },
    {
     "key": "Runs",
     "value": "runs the 125B-class Qwen3.8-Flash-Next model",
     "ev": "official",
     "src": 1
    },
    {
     "key": "License",
     "value": "MIT-licensed inference engine",
     "ev": "official",
     "src": 1
    },
    {
     "key": "Minimum GPU",
     "value": "an NVIDIA GeForce RTX 20–50 series with 8 GB of VRAM (slow; 12 GB recommended)",
     "ev": "official",
     "src": 10
    },
    {
     "key": "Memory",
     "value": "32 GB of RAM and graphics memory (VRAM) together",
     "ev": "official",
     "src": 1
    },
    {
     "key": "Disk",
     "value": "~80 GB free on an SSD",
     "ev": "official",
     "src": 1
    },
    {
     "key": "System",
     "value": "Windows 10/11 or Linux",
     "ev": "official",
     "src": 1
    },
    {
     "key": "Mac",
     "value": "through Strata's Metal engine, ready-made in its Mac app",
     "ev": "official",
     "src": 5
    },
    {
     "key": "Older NVIDIA",
     "value": "experimental and use Strata's separate CUDA 12 engine",
     "ev": "official",
     "src": 6
    }
   ]
  }
 ]
}