{
 "title": "Model Drops",
 "last_updated": "Oct 27, 2026, 9:25 AM ET",
 "window": {
  "from": "2026-09-25",
  "to": "2026-10-25",
  "label": "Sep 25 to Oct 25, 2026"
 },
 "summary": "New AI model releases from Sep 25 to Oct 25, 2026, the quantized builds you can run locally, and trending top 5 lists.",
 "highlights": {
  "lead": "Biggest moves",
  "text": "Kimi K3.1 Pro (Oct 23) scored 55 on the Artificial Analysis Intelligence Index, the highest for a model promising open weights, which are due by Nov 4. DeepSeek, MiniMax, Meta and Reflection all posted 400B+ open weights in mid-October. Decision models became a new category in October, from Perplexity, Liquid AI, TypeSafe, Cloudflare and others."
 },
 "coming_up": "Mistral Large 4 weights (end of Oct), Kimi K3.1 Pro weights (by Nov 4)",
 "drops_caption": "Release date is the vendor's announcement date. Sizes are total / active parameters where published.",
 "quants_caption": "Hugging Face repos confirmed live on Oct 25, 2026. Official means published by the model's own lab.",
 "quants_note": "Kimi K3.1 Pro and Mistral Large 4 have no weights yet, so they have no quants. Muse Spark Open and Beam need hundreds of GB even quantized; Flint, LFM2.5-24B, Nemotron Bolt and Spark-X2.6-4B are the laptop-friendly picks.",
 "probability_note": "What this category means here: probabilistic forecasting and decision models such as TypeSafe AI's Jev (spelled J-e-v) and Perplexity's open pplx-decider models. They answer typed questions with calibrated probabilities for routing, gating and classification. Math-reasoning models are not part of this list.",
 "drops": [
  {
   "name": "Unlimited OCR 2",
   "company": "Baidu",
   "date": "2026-10-27",
   "type": "Open weights",
   "size": "~3.3B",
   "good": "Long multi-page document OCR and parsing; sparse sliding-window attention keeps memory flat on hundred-page docs (MIT license)",
   "url": "https://huggingface.co/baidu/Unlimited-OCR-2"
  },
  {
   "name": "GPT-6.1 Terra",
   "company": "OpenAI",
   "date": "2026-10-26",
   "type": "API",
   "size": "Not disclosed",
   "good": "Fast, low-cost computer use, visual reasoning, writing and coding; 1M context; API prices 50% below GPT-6.1 Sol (vendor)",
   "url": "https://openai.com/index/introducing-gpt-6-1-terra"
  },
  {
   "name": "Claude Sonnet 5.5",
   "company": "Anthropic",
   "date": "2026-09-28",
   "type": "API",
   "size": "Not disclosed",
   "good": "Everyday coding, bug fixing, documents; 70.6% Terminal-Bench 4.0 (vendor)",
   "url": "https://www.anthropic.com/claude-sonnet-5-5"
  },
  {
   "name": "GPT-6.1 Sol",
   "company": "OpenAI",
   "date": "2026-09-29",
   "type": "API",
   "size": "Not disclosed",
   "good": "Near GPT-6 Astra on agentic coding and computer use at a fifth of Astra's price (vendor)",
   "url": "https://openai.com/index/introducing-gpt-6-1-sol"
  },
  {
   "name": "Gemini 4 Argon",
   "company": "Google",
   "date": "2026-10-13",
   "type": "API",
   "size": "Not disclosed",
   "good": "Deep multi-step work in coding, finance, legal, science; #1 Vals Index (vendor). Announced Sep 30, GA Oct 13",
   "url": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-argon/"
  },
  {
   "name": "Mistral Large 4",
   "company": "Mistral AI",
   "date": "2026-10-06",
   "type": "API preview, weights promised",
   "size": "1.05T total / 52B active",
   "good": "Coding, agents, cybersecurity, multimodal; weights due end of October",
   "url": "https://mistral.ai/news/mistral-large-4/"
  },
  {
   "name": "Claude Haiku 5.5",
   "company": "Anthropic",
   "date": "2026-10-07",
   "type": "API",
   "size": "Not disclosed",
   "good": "Cheapest, fastest Claude; subagents, support, browser use",
   "url": "https://www.anthropic.com/claude-haiku-5-5"
  },
  {
   "name": "Muse Wonder / Muse Wonder Full",
   "company": "Meta",
   "date": "2026-10-12",
   "type": "API",
   "size": "Not disclosed",
   "good": "Long-running personal and professional agent tasks",
   "url": "https://about.fb.com/news/2026/10/introducing-muse-wonder-models/"
  },
  {
   "name": "Kimi K3.1 Pro",
   "company": "Moonshot AI",
   "date": "2026-10-23",
   "type": "API, weights by Nov 4",
   "size": "2.8T total / 104B active",
   "good": "Coding agents, finance research, 3D; 55 on AA Intelligence Index",
   "url": "https://www.kimi.ai/blog/kimi-k3-1-pro"
  },
  {
   "name": "MiniMax M3.1 Flash",
   "company": "MiniMax",
   "date": "2026-10-19",
   "type": "Open weights",
   "size": "428B total / 23B active",
   "good": "Cheap agentic workflows, coding, computer use; weights posted Oct 23",
   "url": "https://www.minimax.io/blog/minimax-m31f"
  },
  {
   "name": "DeepSeek V4.1 Flash 1017",
   "company": "DeepSeek",
   "date": "2026-10-17",
   "type": "Open weights",
   "size": "552B backbone (MoE)",
   "good": "Agentic coding at budget prices; MIT licence",
   "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017"
  },
  {
   "name": "Muse Spark Open",
   "company": "Meta",
   "date": "2026-10-15",
   "type": "Open weights",
   "size": "816B total / 42B active",
   "good": "Coding and agents, multimodal, 1M context",
   "url": "https://research.meta.ai/blog/muse-spark-open"
  },
  {
   "name": "Beam 1.0",
   "company": "Reflection AI",
   "date": "2026-10-16",
   "type": "Open weights",
   "size": "501B total / 23B active",
   "good": "General text model, 1M context, Apache 2.0",
   "url": "https://huggingface.co/blog/reflection-beam"
  },
  {
   "name": "Flint",
   "company": "Recursive",
   "date": "2026-10-21",
   "type": "Open weights",
   "size": "40B total / 2.5B active",
   "good": "Fast local agentic coding on a laptop, Apache 2.0",
   "url": "https://recursive.com/updates/introducing-recursive-flint"
  },
  {
   "name": "Nemotron 3.5 Super VL 120B-A12B",
   "company": "NVIDIA",
   "date": "2026-10-15",
   "type": "Open weights",
   "size": "120B total / 12B active",
   "good": "Vision-language",
   "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-BF16"
  },
  {
   "name": "Nemotron 3.5 Bolt 74B-A9B 24GB",
   "company": "NVIDIA",
   "date": "2026-10-15",
   "type": "Open weights",
   "size": "74B total / 9B active",
   "good": "Fits a 24GB GPU (2-bit experts, GGUF)",
   "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB-GGUF"
  },
  {
   "name": "Nemotron 3.5 Bolt E4B 4GB",
   "company": "NVIDIA",
   "date": "2026-10-23",
   "type": "Open weights",
   "size": "E4B",
   "good": "Small on-device model",
   "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Bolt-E4B-4GB"
  },
  {
   "name": "LFM2.5-24B-A2B",
   "company": "Liquid AI",
   "date": "2026-10-13",
   "type": "Open weights",
   "size": "24B total / 2B active",
   "good": "Efficient local model, 131K context",
   "url": "https://huggingface.co/LiquidAI/LFM2.5-24B-A2B"
  },
  {
   "name": "Hy-Memory-30B-A3B",
   "company": "Tencent",
   "date": "2026-10-15",
   "type": "Open weights",
   "size": "30B total / 3B active",
   "good": "Memory writer for agent memory systems",
   "url": "https://huggingface.co/tencent/Hy-Memory-30B-A3B"
  },
  {
   "name": "Ling 3.1 Flash",
   "company": "InclusionAI",
   "date": "2026-10-16",
   "type": "Open weights",
   "size": "560B total / 25B active",
   "good": "Long-horizon coding, MIT",
   "url": "https://huggingface.co/inclusionAI/Ling-3.1-flash"
  },
  {
   "name": "Sage-A1.5",
   "company": "InternScience",
   "date": "2026-10-15",
   "type": "Open weights",
   "size": "36B",
   "good": "Trending on Hugging Face; see model card",
   "url": "https://huggingface.co/InternScience/Sage-A1.5"
  },
  {
   "name": "Spark-X2.6-4B",
   "company": "SparkLLM (XHToken)",
   "date": "2026-10-19",
   "type": "Open weights",
   "size": "4B",
   "good": "On-device model, Apache 2.0",
   "url": "https://huggingface.co/XHToken/Spark-X2.6-4B"
  },
  {
   "name": "Granite Switch 4.1 (8B, 30B)",
   "company": "IBM",
   "date": "2026-10-17",
   "type": "Open weights",
   "size": "8B, 30B",
   "good": "Bundled decision and guardian adapters, Apache 2.0",
   "url": "https://huggingface.co/ibm-granite/granite-switch-4.1-8b"
  },
  {
   "name": "Jais 3 family",
   "company": "IFM",
   "date": "2026-10-15",
   "type": "Open weights",
   "size": "0.9B to 375B-A23B",
   "good": "Arabic and English family in six sizes",
   "url": "https://huggingface.co/inception42/Jais-3-32B"
  },
  {
   "name": "Mellum2.1-12B-A2.5B-Thinking",
   "company": "JetBrains",
   "date": "2026-10-08",
   "type": "Open weights",
   "size": "12B total / 2.5B active",
   "good": "Code reasoning, Apache 2.0",
   "url": "https://huggingface.co/JetBrains/Mellum2.1-12B-A2.5B-Thinking"
  },
  {
   "name": "Nanbeige4.3-33B",
   "company": "Nanbeige",
   "date": "2026-10-24",
   "type": "Open weights",
   "size": "33B",
   "good": "Newest upload in this list; see model card",
   "url": "https://huggingface.co/Nanbeige/Nanbeige4.3-33B"
  },
  {
   "name": "Index-Translate-35B-A3B-preview",
   "company": "Bilibili Index team",
   "date": "2026-10-02",
   "type": "Open weights",
   "size": "35B total / 3B active",
   "good": "Translation across 150 languages, Apache 2.0",
   "url": "https://huggingface.co/IndexTeam/Index-Translate-35B-A3B-preview"
  },
  {
   "name": "Voxtral 3",
   "company": "Mistral AI",
   "date": "2026-10-20",
   "type": "API",
   "size": "Small and Mini variants",
   "good": "Speech transcription and text-to-speech",
   "url": "https://benchlm.ai/model-updates/releases/october-2026"
  },
  {
   "name": "Mercury 3",
   "company": "Inception",
   "date": "2026-10-19",
   "type": "API",
   "size": "Not disclosed",
   "good": "Inception's new Mercury generation",
   "url": "https://benchlm.ai/model-updates/releases/october-2026"
  },
  {
   "name": "Clef / Clef-Flash",
   "company": "Cloudflare",
   "date": "2026-10-01",
   "type": "Open weights",
   "size": "27B, 9B",
   "good": "Decision model: probabilities for routing and checks, Apache 2.0",
   "url": "https://huggingface.co/Cloudflare/clef"
  },
  {
   "name": "pplx-decider-v1.2 / Flash v1.2",
   "company": "Perplexity",
   "date": "2026-10-16",
   "type": "Open weights",
   "size": "27B, 4B",
   "good": "Decision model: yes/no, choice, rubric probabilities; text and images",
   "url": "https://huggingface.co/perplexity-ai/pplx-decider-v1.2"
  },
  {
   "name": "Jev 1.14",
   "company": "TypeSafe AI",
   "date": "2026-10-14",
   "type": "API",
   "size": "Not disclosed",
   "good": "System One decision model: calibrated choice, score, yes/no",
   "url": "https://typesafe.ai/blog/jev-1-14"
  },
  {
   "name": "d1-ultra",
   "company": "Liquid AI",
   "date": "2026-10-15",
   "type": "API",
   "size": "Not disclosed",
   "good": "Multimodal decision model, calibrated probabilities",
   "url": "https://www.liquid.ai/blog/d1-ultra"
  },
  {
   "name": "d1-3B / d1-omni-600M",
   "company": "Liquid AI",
   "date": "2026-10-07",
   "type": "Open weights",
   "size": "3.1B, 587M",
   "good": "Small local decision models",
   "url": "https://docs.liquid.ai/lfm/models/decision-models"
  },
  {
   "name": "Microsoft-Decision-1",
   "company": "Microsoft",
   "date": "2026-10-12",
   "type": "API",
   "size": "Not disclosed",
   "good": "Decision model",
   "url": "https://benchlm.ai/model-updates/releases/october-2026"
  },
  {
   "name": "Drex Max",
   "company": "Nace.AI",
   "date": "2026-10-19",
   "type": "API",
   "size": "Not disclosed",
   "good": "Multimodal decision model for computer and browser use",
   "url": "https://console.nace.ai/docs/drex/guides/drex-max"
  },
  {
   "name": "LingDT-Decision-1.0 (4B, 27B)",
   "company": "Ant Digital Technologies",
   "date": "2026-10-22",
   "type": "API",
   "size": "4B, 27B",
   "good": "Decision model with native multi-select and ordered (DAG) decisions; vendor says 27B beats Microsoft-Decision-1 and Jev 1.13 (vendor)",
   "url": "https://maas.antdigital.com/docs/about/lingdt-decision-release"
  },
  {
   "name": "Carbon-X",
   "company": "Hugging Face",
   "date": "2026-10-25",
   "type": "Open weights",
   "size": "14.8B",
   "good": "Reads raw DNA in chat and answers in English (genomics)",
   "url": "https://huggingface.co/HuggingFaceBio/Carbon-X"
  },
  {
   "name": "Kolibri 1",
   "company": "Aleph Alpha",
   "date": "2026-10-02",
   "type": "Open weights",
   "size": "78B total / 3.46B active",
   "good": "German and English MoE reasoning model with tool calling, 1M context",
   "url": "https://aleph-alpha.com/en/blog/kolibri-has-landed-a-sovereign-open-weight-model/"
  }
 ],
 "quants": {
  "Unlimited OCR 2": [
   {
    "format": "MLX 4-bit",
    "publisher": "mlx-community",
    "repo": "mlx-community/Unlimited-OCR-2-4bit"
   },
   {
    "format": "MLX 8-bit",
    "publisher": "mlx-community",
    "repo": "mlx-community/Unlimited-OCR-2-8bit"
   },
   {
    "format": "GGUF",
    "publisher": "AtomicChat",
    "repo": "AtomicChat/Unlimited-OCR-2-GGUF"
   },
   {
    "format": "W4A16",
    "publisher": "AtomicChat",
    "repo": "AtomicChat/Unlimited-OCR-2-W4A16"
   }
  ],
  "DeepSeek V4.1 Flash 1017": [
   {
    "format": "GGUF",
    "publisher": "unsloth",
    "repo": "unsloth/DeepSeek-V4.1-Flash-1017-GGUF"
   },
   {
    "format": "GGUF (GSQ-RCO)",
    "publisher": "ISTA-DASLab",
    "repo": "ISTA-DASLab/DeepSeek-V4.1-Flash-1017-GSQ-RCO-GGUF"
   },
   {
    "format": "MLX 4-bit",
    "publisher": "mlx-community",
    "repo": "mlx-community/DeepSeek-V4.1-Flash-1017-oQ4e-mtp"
   },
   {
    "format": "EXL3 2.9bpw",
    "publisher": "LibertAIDAI",
    "repo": "LibertAIDAI/DeepSeek-V4.1-Flash-1017-EXL3-2.9bpw"
   }
  ],
  "MiniMax M3.1 Flash": [
   {
    "format": "GGUF",
    "publisher": "unsloth",
    "repo": "unsloth/MiniMax-M3.1-Flash-GGUF"
   },
   {
    "format": "MLX 4-bit",
    "publisher": "mlx-community",
    "repo": "mlx-community/MiniMax-M3.1-Flash-Text-4bit"
   },
   {
    "format": "MXFP4 (Quark)",
    "publisher": "amd",
    "repo": "amd/MiniMax-M3.1-Flash-Quark-MXFP4"
   },
   {
    "format": "EXL3 3.06bpw",
    "publisher": "Mia-AiLab",
    "repo": "Mia-AiLab/MiniMax-M3.1-Flash-EXL3-EAQ-v1-3.06-bpw"
   }
  ],
  "Muse Spark Open": [
   {
    "format": "NVFP4 (official)",
    "publisher": "meta-models",
    "repo": "meta-models/Muse-Spark-Open-816B-A42B-NVFP4"
   },
   {
    "format": "GGUF",
    "publisher": "unsloth",
    "repo": "unsloth/Muse-Spark-Open-816B-A42B-GGUF"
   },
   {
    "format": "GGUF",
    "publisher": "ggml-org",
    "repo": "ggml-org/Muse-Spark-Open-816B-A42B-GGUF"
   },
   {
    "format": "MLX 4-bit",
    "publisher": "mlx-community",
    "repo": "mlx-community/Muse-Spark-Open-816B-A42B-4bit"
   },
   {
    "format": "INT4",
    "publisher": "intel-ai",
    "repo": "intel-ai/Muse-Spark-Open-816B-A42B-INT4"
   }
  ],
  "Beam 1.0": [
   {
    "format": "FP8 (official)",
    "publisher": "reflection",
    "repo": "reflection/Beam-1.0-501B-A23B-FP8"
   },
   {
    "format": "NVFP4 (official)",
    "publisher": "reflection",
    "repo": "reflection/Beam-1.0-501B-A23B-NVFP4"
   },
   {
    "format": "GGUF",
    "publisher": "unsloth",
    "repo": "unsloth/Beam-1.0-501B-A23B-GGUF"
   },
   {
    "format": "GGUF",
    "publisher": "bartowski",
    "repo": "bartowski/Beam-1.0-501B-A23B-GGUF"
   },
   {
    "format": "EXL3 3bpw",
    "publisher": "diffbot",
    "repo": "diffbot/Beam-1.0-501B-A23B-EXL3-3bpw-2x-RTX-PRO-6000"
   }
  ],
  "Flint": [
   {
    "format": "INT4 GGUF (official)",
    "publisher": "RecursiveSI",
    "repo": "RecursiveSI/flint-int4-gguf"
   },
   {
    "format": "GGUF",
    "publisher": "unsloth",
    "repo": "unsloth/flint-GGUF"
   },
   {
    "format": "GGUF",
    "publisher": "lmstudio-community",
    "repo": "lmstudio-community/flint-GGUF"
   },
   {
    "format": "GGUF",
    "publisher": "bartowski",
    "repo": "bartowski/RecursiveSI_flint-GGUF"
   },
   {
    "format": "MLX 3/4-bit",
    "publisher": "mlx-community",
    "repo": "mlx-community/flint-mixed-3-4bit"
   },
   {
    "format": "MLX GPTQ MXFP4",
    "publisher": "RESMP-DEV",
    "repo": "RESMP-DEV/Flint-40B-A2.5B-MLX-GPTQ-MXFP4"
   }
  ],
  "Nemotron 3.5 Super VL 120B-A12B": [
   {
    "format": "NVFP4 (official)",
    "publisher": "nvidia",
    "repo": "nvidia/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-NVFP4"
   },
   {
    "format": "GGUF",
    "publisher": "unsloth",
    "repo": "unsloth/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-GGUF"
   },
   {
    "format": "GGUF",
    "publisher": "lmstudio-community",
    "repo": "lmstudio-community/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-GGUF"
   }
  ],
  "Nemotron 3.5 Bolt 74B-A9B": [
   {
    "format": "GGUF (official)",
    "publisher": "nvidia",
    "repo": "nvidia/NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB-GGUF"
   },
   {
    "format": "GGUF",
    "publisher": "unsloth",
    "repo": "unsloth/NVIDIA-Nemotron-3.5-Bolt-74B-A9B-GGUF"
   }
  ],
  "Nemotron 3.5 Bolt E4B": [
   {
    "format": "GGUF",
    "publisher": "unsloth",
    "repo": "unsloth/NVIDIA-Nemotron-3.5-Bolt-E4B-GGUF"
   }
  ],
  "LFM2.5-24B-A2B": [
   {
    "format": "GGUF (official)",
    "publisher": "LiquidAI",
    "repo": "LiquidAI/LFM2.5-24B-A2B-GGUF"
   },
   {
    "format": "MLX 4-bit (official)",
    "publisher": "LiquidAI",
    "repo": "LiquidAI/LFM2.5-24B-A2B-MLX-4bit"
   },
   {
    "format": "GGUF",
    "publisher": "unsloth",
    "repo": "unsloth/LFM2.5-24B-A2B-GGUF"
   },
   {
    "format": "GGUF",
    "publisher": "bartowski",
    "repo": "bartowski/LFM2.5-24B-A2B-GGUF"
   }
  ],
  "Hy-Memory-30B-A3B": [
   {
    "format": "GGUF (official)",
    "publisher": "tencent",
    "repo": "tencent/Hy-Memory-30B-A3B-GGUF"
   },
   {
    "format": "FP8 (official)",
    "publisher": "tencent",
    "repo": "tencent/Hy-Memory-30B-A3B-FP8"
   },
   {
    "format": "EXL3 4.08bpw",
    "publisher": "jyohukuchan",
    "repo": "jyohukuchan/Hy-Memory-30B-A3B-exl3-4.08bpw"
   }
  ],
  "Ling 3.1 Flash": [
   {
    "format": "GGUF",
    "publisher": "bartowski",
    "repo": "bartowski/Ling-3.1-flash-GGUF"
   },
   {
    "format": "EXL3",
    "publisher": "turboderp",
    "repo": "turboderp/Ling-3.1-flash-exl3"
   }
  ],
  "Mellum2.1-12B-A2.5B-Thinking": [
   {
    "format": "AWQ INT4",
    "publisher": "cyankiwi",
    "repo": "cyankiwi/Mellum2.1-12B-A2.5B-Thinking-AWQ-INT4"
   },
   {
    "format": "MLX 6-bit",
    "publisher": "mlx-community",
    "repo": "mlx-community/Mellum2.1-12B-A2.5B-Thinking-6bit"
   },
   {
    "format": "GGUF",
    "publisher": "prithivMLmods",
    "repo": "prithivMLmods/JetBrains-Mellum2.1-12B-A2.5B-Thinking-GGUF"
   }
  ],
  "Spark-X2.6-4B": [
   {
    "format": "GGUF (official)",
    "publisher": "XHToken",
    "repo": "XHToken/Spark-X2.6-4B-GGUF"
   },
   {
    "format": "GGUF",
    "publisher": "bartowski",
    "repo": "bartowski/Spark-X2.6-4B-GGUF"
   }
  ],
  "Nanbeige4.3-33B": [
   {
    "format": "GGUF",
    "publisher": "bartowski",
    "repo": "bartowski/Nanbeige_Nanbeige4.3-33B-GGUF"
   },
   {
    "format": "MLX 4-bit",
    "publisher": "Duckets",
    "repo": "Duckets/Nanbeige4.3-33B-MLX-4bit"
   }
  ],
  "Clef / Clef-Flash": [
   {
    "format": "GGUF",
    "publisher": "bartowski",
    "repo": "bartowski/Cloudflare_clef-GGUF"
   },
   {
    "format": "GGUF",
    "publisher": "bartowski",
    "repo": "bartowski/Cloudflare_clef-flash-GGUF"
   },
   {
    "format": "GGUF",
    "publisher": "ggml-org",
    "repo": "ggml-org/Clef-Flash-GGUF"
   },
   {
    "format": "MLX 4-bit",
    "publisher": "mlx-community",
    "repo": "mlx-community/clef-4bit"
   }
  ],
  "Carbon-X": [
   {
    "format": "GGUF (official)",
    "publisher": "HuggingFaceBio",
    "repo": "HuggingFaceBio/Carbon-X-GGUF"
   },
   {
    "format": "GGUF",
    "publisher": "mradermacher",
    "repo": "mradermacher/Carbon-X-GGUF"
   },
   {
    "format": "GGUF (imatrix)",
    "publisher": "mradermacher",
    "repo": "mradermacher/Carbon-X-i1-GGUF"
   }
  ],
  "Kolibri 1": [
   {
    "format": "BF16 (official)",
    "publisher": "Aleph-Alpha",
    "repo": "Aleph-Alpha/Kolibri-1-BF16"
   },
   {
    "format": "GGUF",
    "publisher": "bartowski",
    "repo": "bartowski/Kolibri-1-GGUF"
   },
   {
    "format": "MLX 4-bit",
    "publisher": "velaia",
    "repo": "velaia/Kolibri-1-MLX-4bit"
   }
  ],
  "pplx-decider-flash-v1.2": [
   {
    "format": "GGUF (official)",
    "publisher": "perplexity-ai",
    "repo": "perplexity-ai/pplx-decider-flash-gguf-v1.2"
   }
  ]
 },
 "boards": [
  {
   "key": "coding",
   "label": "Coding",
   "source": "LMArena WebDev leaderboard",
   "url": "https://arena.ai/leaderboard/webdev",
   "snapshot": "Oct 22, 2026",
   "note": "Ranked by Arena WebDev score (human votes on web-dev and agentic coding tasks). Cross-check: Artificial Analysis Coding Agent Index v2.0 (Oct 23) has Opus 5.5 at 61.7, Sonnet 5.5 at 59.3, Muse Wonder Full at 58.9.",
   "rows": [
    {
     "rank": 1,
     "model": "Claude Opus 5.5 (Max)",
     "company": "Anthropic",
     "score": "1807",
     "why": "Leads WebDev by 33 points and tops the AA Coding Agent Index"
    },
    {
     "rank": 2,
     "model": "GPT-6 Astra (Max)",
     "company": "OpenAI",
     "score": "1774",
     "why": "OpenAI's strongest coder; pricey at $10 / $50 per M tokens"
    },
    {
     "rank": 3,
     "model": "Claude Sonnet 5.5 (xHigh)",
     "company": "Anthropic",
     "score": "1769",
     "why": "Near-Opus coding at $2 / $10; #2 on AA Coding Agent Index"
    },
    {
     "rank": 4,
     "model": "GPT-6.1 Sol (Max)",
     "company": "OpenAI",
     "score": "1749",
     "why": "On AA's cost frontier: 54.1 index at about $1.50 per task"
    },
    {
     "rank": 5,
     "model": "Claude Fable 5.1 (Max)",
     "company": "Anthropic",
     "score": "1742",
     "why": "Top-5 WebDev and 57.6 on AA Coding Agent Index"
    }
   ]
  },
  {
   "key": "agentic",
   "label": "Agentic / tool use",
   "source": "Artificial Analysis GDPval-AA v2.1",
   "url": "https://artificialanalysis.ai/evaluations/gdpval-aa",
   "snapshot": "read Oct 25, 2026 (latest result added Oct 24)",
   "note": "GDPval-AA scores agents on 220 real knowledge-work tasks with tools, judged pairwise into Elo. Distinct models only; best setting per model.",
   "rows": [
    {
     "rank": 1,
     "model": "Claude Opus 5.5 (Max)",
     "company": "Anthropic",
     "score": "1866 Elo",
     "why": "Best real-world agent deliverables measured"
    },
    {
     "rank": 2,
     "model": "Muse Wonder Full (Max)",
     "company": "Meta",
     "score": "1857 Elo",
     "why": "Meta's agent-first model, Muse Code only"
    },
    {
     "rank": 3,
     "model": "Kimi K3.1 Pro",
     "company": "Moonshot AI",
     "score": "1846 Elo",
     "why": "Top open-weight-bound agent; 67% Terminal-Bench 4.0 per AA"
    },
    {
     "rank": 4,
     "model": "Claude Sonnet 5.5 (Max)",
     "company": "Anthropic",
     "score": "1838 Elo",
     "why": "Near the top at a fraction of Opus pricing"
    },
    {
     "rank": 5,
     "model": "Muse Wonder (Max)",
     "company": "Meta",
     "score": "1770 Elo",
     "why": "API-available sibling of Wonder Full"
    }
   ]
  },
  {
   "key": "probability",
   "label": "Probability / decision",
   "source": "DecisionBench overall ranking",
   "url": "https://www.decisionbench.lol/",
   "snapshot": "Oct 22, 2026",
   "note": "Decision models return calibrated probabilities (yes/no, choice, rubric score) instead of text. DecisionBench is an independent board of 18 OpenRouter decision models across five tests: Sudoku, Rulebook, Consistency, Odds and Fairness.",
   "rows": [
    {
     "rank": 1,
     "model": "d1-ultra",
     "company": "Liquid AI",
     "score": "90.4",
     "why": "Released Oct 15; tier 1 alone, top-3 on four of five tests"
    },
    {
     "rank": 2,
     "model": "Drex Max",
     "company": "Nace.AI",
     "score": "83.7",
     "why": "Multimodal, built for computer and browser use; #1 Rulebook and Fairness"
    },
    {
     "rank": 3,
     "model": "pplx-decider-v1.2",
     "company": "Perplexity",
     "score": "75.5",
     "why": "Open weights (27B); perfect Consistency; tops the Decision Index at 68.0"
    },
    {
     "rank": 4,
     "model": "Jev 1.14",
     "company": "TypeSafe AI",
     "score": "69.3",
     "why": "The original System One decision model, trained for calibration"
    },
    {
     "rank": 5,
     "model": "Clef",
     "company": "Cloudflare",
     "score": "62.6",
     "why": "Apache 2.0 open weights (27B), cheap on Workers AI"
    }
   ]
  },
  {
   "key": "prose",
   "label": "Prose / creative writing",
   "source": "LMArena Text: Creative Writing",
   "url": "https://arena.ai/leaderboard/chat/text/creative-writing",
   "snapshot": "Oct 22, 2026",
   "note": "Ranked by Arena creative-writing score from human votes. Gemini 3.7 Flash (High) ties Muse Wonder at 1492 but is marked preliminary.",
   "rows": [
    {
     "rank": 1,
     "model": "Claude Opus 5.5 (High)",
     "company": "Anthropic",
     "score": "1520",
     "why": "Readers' top pick for creative writing"
    },
    {
     "rank": 2,
     "model": "Gemini 4 Argon (High)",
     "company": "Google",
     "score": "1512",
     "why": "Close second, at half Opus pricing"
    },
    {
     "rank": 3,
     "model": "Claude Fable 5 (High)",
     "company": "Anthropic",
     "score": "1501",
     "why": "Most-voted of the top 3 (9,586 votes)"
    },
    {
     "rank": 4,
     "model": "Claude Opus 4.6 (High)",
     "company": "Anthropic",
     "score": "1499",
     "why": "Older model still holding a top-5 spot"
    },
    {
     "rank": 5,
     "model": "Muse Wonder (Max)",
     "company": "Meta",
     "score": "1492",
     "why": "Meta's first top-5 creative-writing entry"
    }
   ]
  }
 ],
 "sources": [
  {
   "title": "AI model releases in October 2026",
   "site": "BenchLM",
   "href": "https://benchlm.ai/model-updates/releases/october-2026"
  },
  {
   "title": "Hugging Face trending models",
   "site": "Hugging Face",
   "href": "https://huggingface.co/models?sort=trending"
  },
  {
   "title": "Kimi K3.1 Pro launch benchmarks",
   "site": "Artificial Analysis",
   "date": "Oct 23, 2026",
   "href": "https://artificialanalysis.ai/articles/kimi-k3-1-pro-launch-benchmarks"
  },
  {
   "title": "WebDev leaderboard",
   "site": "LMArena",
   "date": "Oct 22, 2026",
   "href": "https://arena.ai/leaderboard/webdev"
  },
  {
   "title": "Creative Writing leaderboard",
   "site": "LMArena",
   "date": "Oct 22, 2026",
   "href": "https://arena.ai/leaderboard/chat/text/creative-writing"
  },
  {
   "title": "GDPval-AA v2.1",
   "site": "Artificial Analysis",
   "href": "https://artificialanalysis.ai/evaluations/gdpval-aa"
  },
  {
   "title": "AI Decision Model Leaderboard",
   "site": "DecisionBench",
   "date": "Oct 22, 2026",
   "href": "https://www.decisionbench.lol/"
  },
  {
   "title": "Coding Agent Index",
   "site": "Artificial Analysis",
   "href": "https://artificialanalysis.ai/agents/coding-agents"
  }
 ],
 "sources_note": "Not verified on a vendor page: exact dates and details for Voxtral 3, Mercury 3 and Microsoft-Decision-1 come from the BenchLM release tracker. The Drex Max date comes from MarkTechPost coverage (Oct 19). DeepSeek's 552B size comes from release coverage. Vendor benchmark claims are marked \"(vendor)\". EQ-Bench Creative Writing could not be read directly, so prose rankings use LMArena only.",
 "generated_at": "2026-10-27T14:03:33.635Z"
}