{"generated":"2026-10-15T18:13:06Z","window_days":30,"status":{"hf_candidates":76,"hf_models":76,"github_fail":[],"github_ok":15,"ollama":252,"x_posts":47},"models":[{"type":"model","id":"nvidia/Nemotron-3.5-Rerank-8B-BF16","name":"Nemotron-3.5-Rerank-8B-BF16","org":"nvidia","date":"2026-10-15","license":"openmdw-1.1","params":8381159424,"params_src":"safetensors","active":null,"ctx":262144,"kv_bytes_tok":139264,"dtypes":["BF16"],"pipeline":"text-ranking","modality":"embedding","summary":"This model is ready for commercial or non-commercial use.","url":"https://huggingface.co/nvidia/Nemotron-3.5-Rerank-8B-BF16","likes":0,"downloads":439,"gated":false,"quants":[],"x_mentions":[],"ollama":"https://ollama.com/library/nemotron3","est":{"weights":{"Q4":5.0,"Q8":8.9,"FP16":16.8},"kv_gb":1.1,"kv_src":"config","need":{"Q4":7.1,"Q8":11.0,"FP16":18.9},"fit":{"gpu12":"Q8","gpu16":"Q8","gpu24":"FP16","gpu32":"FP16","gpu2x24":"FP16","spark1":"FP16","spark2":"FP16","mac32":"FP16","mac64":"FP16","mac128":"FP16","mac192":"FP16","mac512":"FP16","cpu":"FP16"}}},{"type":"model","id":"black-forest-labs/flux-3-image-klein-comfy","name":"flux-3-image-klein-comfy","org":"black-forest-labs","date":"2026-10-14","license":"flux-kommunity-license","params":null,"params_src":null,"active":null,"ctx":null,"kv_bytes_tok":null,"dtypes":[],"pipeline":"image-to-image","modality":"image","summary":"FLUX 3 Image Klein is a 5.6B open weights model for text-to-image generation and image editing.","url":"https://huggingface.co/black-forest-labs/flux-3-image-klein-comfy","likes":104,"downloads":39,"gated":true,"quants":[],"x_mentions":[],"est":null},{"type":"model","id":"mistralai/Timestral-209M","name":"Timestral-209M","org":"mistralai","date":"2026-10-14","license":"other","params":209312256,"params_src":"safetensors","active":null,"ctx":null,"kv_bytes_tok":null,"dtypes":["F32"],"pipeline":"time-series-forecasting","modality":"other","summary":"Timestral is a family of time series foundation models for zero-shot forecasting, available in 5M, 28M, 123M, and 209M parameter versions.","url":"https://huggingface.co/mistralai/Timestral-209M","likes":20,"downloads":11,"gated":false,"quants":[],"x_mentions":[],"est":{"weights":{"Q4":0.1,"Q8":0.2,"FP16":0.4},"kv_gb":0.5,"kv_src":"rough","need":{"Q4":1.6,"Q8":1.7,"FP16":1.9},"fit":{"gpu12":"FP16","gpu16":"FP16","gpu24":"FP16","gpu32":"FP16","gpu2x24":"FP16","spark1":"FP16","spark2":"FP16","mac32":"FP16","mac64":"FP16","mac128":"FP16","mac192":"FP16","mac512":"FP16","cpu":"FP16"}}},{"type":"model","id":"mistralai/Timestral-28M","name":"Timestral-28M","org":"mistralai","date":"2026-10-14","license":"other","params":27821184,"params_src":"safetensors","active":null,"ctx":null,"kv_bytes_tok":null,"dtypes":["F32"],"pipeline":"time-series-forecasting","modality":"other","summary":"Timestral is a family of time series foundation models for zero-shot forecasting, available in 5M, 28M, 123M, and 209M parameter versions.","url":"https://huggingface.co/mistralai/Timestral-28M","likes":12,"downloads":5,"gated":false,"quants":[],"x_mentions":[],"est":{"weights":{"Q4":0.0,"Q8":0.0,"FP16":0.1},"kv_gb":0.5,"kv_src":"rough","need":{"Q4":1.5,"Q8":1.5,"FP16":1.6},"fit":{"gpu12":"FP16","gpu16":"FP16","gpu24":"FP16","gpu32":"FP16","gpu2x24":"FP16","spark1":"FP16","spark2":"FP16","mac32":"FP16","mac64":"FP16","mac128":"FP16","mac192":"FP16","mac512":"FP16","cpu":"FP16"}}},{"type":"model","id":"mistralai/Timestral-5M","name":"Timestral-5M","org":"mistralai","date":"2026-10-14","license":"other","params":5208960,"params_src":"safetensors","active":null,"ctx":null,"kv_bytes_tok":null,"dtypes":["F32"],"pipeline":"time-series-forecasting","modality":"other","summary":"Timestral is a family of time series foundation models for zero-shot forecasting, available in 5M, 28M, 123M, and 209M parameter versions.","url":"https://huggingface.co/mistralai/Timestral-5M","likes":12,"downloads":7,"gated":false,"quants":[],"x_mentions":[],"est":{"weights":{"Q4":0.0,"Q8":0.0,"FP16":0.0},"kv_gb":0.5,"kv_src":"rough","need":{"Q4":1.5,"Q8":1.5,"FP16":1.5},"fit":{"gpu12":"FP16","gpu16":"FP16","gpu24":"FP16","gpu32":"FP16","gpu2x24":"FP16","spark1":"FP16","spark2":"FP16","mac32":"FP16","mac64":"FP16","mac128":"FP16","mac192":"FP16","mac512":"FP16","cpu":"FP16"}}},{"type":"model","id":"mistralai/Timestral-123M","name":"Timestral-123M","org":"mistralai","date":"2026-10-14","license":"other","params":122606720,"params_src":"safetensors","active":null,"ctx":null,"kv_bytes_tok":null,"dtypes":["F32"],"pipeline":"time-series-forecasting","modality":"other","summary":"Timestral is a family of time series foundation models for zero-shot forecasting, available in 5M, 28M, 123M, and 209M parameter versions.","url":"https://huggingface.co/mistralai/Timestral-123M","likes":11,"downloads":5,"gated":false,"quants":[],"x_mentions":[],"est":{"weights":{"Q4":0.1,"Q8":0.1,"FP16":0.2},"kv_gb":0.5,"kv_src":"rough","need":{"Q4":1.6,"Q8":1.6,"FP16":1.7},"fit":{"gpu12":"FP16","gpu16":"FP16","gpu24":"FP16","gpu32":"FP16","gpu2x24":"FP16","spark1":"FP16","spark2":"FP16","mac32":"FP16","mac64":"FP16","mac128":"FP16","mac192":"FP16","mac512":"FP16","cpu":"FP16"}}},{"type":"model","id":"microsoft/UserRepGen-4B","name":"UserRepGen-4B","org":"microsoft","date":"2026-10-14","license":"mit","params":4539265536,"params_src":"safetensors","active":null,"ctx":262144,"kv_bytes_tok":131072,"dtypes":["BF16"],"pipeline":"text-generation","modality":"text","summary":"UserRepGen-4B is a fine-tuned Qwen3.5-4B language model that generates a structured, natural-language user representation from a multi-turn user-assistant conversation.","url":"https://huggingface.co/microsoft/UserRepGen-4B","likes":6,"downloads":481,"gated":false,"quants":[{"id":"bartowski/UserRepGen-4B-GGUF","kind":"GGUF"}],"x_mentions":[],"est":{"weights":{"Q4":2.7,"Q8":4.8,"FP16":9.1},"kv_gb":1.1,"kv_src":"config","need":{"Q4":4.8,"Q8":6.9,"FP16":11.2},"fit":{"gpu12":"Q8","gpu16":"FP16","gpu24":"FP16","gpu32":"FP16","gpu2x24":"FP16","spark1":"FP16","spark2":"FP16","mac32":"FP16","mac64":"FP16","mac128":"FP16","mac192":"FP16","mac512":"FP16","cpu":"FP16"}}},{"type":"model","id":"black-forest-labs/flux-3-image-klein","name":"flux-3-image-klein","org":"black-forest-labs","date":"2026-10-13","license":"flux-kommunity-license","params":5548681984,"params_src":"safetensors","active":null,"ctx":null,"kv_bytes_tok":null,"dtypes":["BF16"],"pipeline":"image-to-image","modality":"image","summary":"FLUX 3 Image Klein is a 5.6B open weights model for text-to-image generation and image editing.","url":"https://huggingface.co/black-forest-labs/flux-3-image-klein","likes":128,"downloads":3241,"gated":true,"quants":[],"x_mentions":[],"est":{"weights":{"Q4":3.3,"Q8":5.9,"FP16":11.1},"kv_gb":0.5,"kv_src":"rough","need":{"Q4":4.8,"Q8":7.4,"FP16":12.6},"fit":{"gpu12":"Q8","gpu16":"FP16","gpu24":"FP16","gpu32":"FP16","gpu2x24":"FP16","spark1":"FP16","spark2":"FP16","mac32":"FP16","mac64":"FP16","mac128":"FP16","mac192":"FP16","mac512":"FP16","cpu":"FP16"}}},{"type":"model","id":"inclusionAI/Ling-3.1-flash","name":"Ling-3.1-flash","org":"inclusionAI","date":"2026-10-13","license":"mit","params":572120680768,"params_src":"safetensors","active":null,"ctx":262144,"kv_bytes_tok":82944,"dtypes":["BF16","F8_E4M3","F32"],"pipeline":"text-generation","modality":"text","summary":"With stronger real-world capabilities, Ling-3.1-flash shows improvements across a range of real-world tasks, including general-purpose agents, search, office productivity, and software engineering.","url":"https://huggingface.co/inclusionAI/Ling-3.1-flash","likes":101,"downloads":666,"gated":false,"quants":[],"x_mentions":[],"est":{"weights":{"Q4":343.3,"Q8":607.9,"FP16":1144.2},"kv_gb":0.7,"kv_src":"config","need":{"Q4":345.0,"Q8":609.6,"FP16":1145.9},"fit":{"gpu12":null,"gpu16":null,"gpu24":null,"gpu32":null,"gpu2x24":null,"spark1":null,"spark2":null,"mac32":null,"mac64":null,"mac128":null,"mac192":null,"mac512":"Q4","cpu":null}}},{"type":"model","id":"black-forest-labs/flux-3-image-klein-base","name":"flux-3-image-klein-base","org":"black-forest-labs","date":"2026-10-13","license":"flux-kommunity-license","params":5538458368,"params_src":"safetensors","active":null,"ctx":null,"kv_bytes_tok":null,"dtypes":["BF16"],"pipeline":"image-to-image","modality":"image","summary":"FLUX 3 Image Klein is a 5.6B open weights model for text-to-image generation and image editing.","url":"https://huggingface.co/black-forest-labs/flux-3-image-klein-base","likes":38,"downloads":134,"gated":true,"quants":[],"x_mentions":[],"est":{"weights":{"Q4":3.3,"Q8":5.9,"FP16":11.1},"kv_gb":0.5,"kv_src":"rough","need":{"Q4":4.8,"Q8":7.4,"FP16":12.6},"fit":{"gpu12":"Q8","gpu16":"FP16","gpu24":"FP16","gpu32":"FP16","gpu2x24":"FP16","spark1":"FP16","spark2":"FP16","mac32":"FP16","mac64":"FP16","mac128":"FP16","mac192":"FP16","mac512":"FP16","cpu":"FP16"}}},{"type":"model","id":"LiquidAI/LFM2.5-24B-A2B","name":"LFM2.5-24B-A2B","org":"LiquidAI","date":"2026-10-12","license":"lfm1.0","params":23971585280,"params_src":"safetensors","active":2000000000.0,"ctx":131072,"kv_bytes_tok":81920,"dtypes":["F32","BF16"],"pipeline":"text-generation","modality":"text","summary":"LFM2.5-24B-A2B is a 24B-parameter reasoning model with only 2B active parameters, built for fast, private agents on-device.","url":"https://huggingface.co/LiquidAI/LFM2.5-24B-A2B","likes":379,"downloads":6286,"gated":false,"quants":[{"id":"AtomicChat/LFM2.5-24B-A2B-Uncensored-GGUF","kind":"GGUF"},{"id":"LiquidAI/LFM2.5-24B-A2B-FP8","kind":"quant"},{"id":"LiquidAI/LFM2.5-24B-A2B-GGUF","kind":"GGUF"},{"id":"LiquidAI/LFM2.5-24B-A2B-MLX-4bit","kind":"MLX"},{"id":"LiquidAI/LFM2.5-24B-A2B-MLX-5bit","kind":"MLX"},{"id":"LiquidAI/LFM2.5-24B-A2B-MLX-6bit","kind":"MLX"},{"id":"LiquidAI/LFM2.5-24B-A2B-MLX-8bit","kind":"MLX"},{"id":"LiquidAI/LFM2.5-24B-A2B-MLX-bf16","kind":"MLX"},{"id":"LiquidAI/LFM2.5-24B-A2B-ONNX","kind":"quant"},{"id":"bartowski/LiquidAI_LFM2.5-24B-A2B-GGUF","kind":"GGUF"},{"id":"ggml-org/LFM2.5-24B-A2B-GGUF","kind":"GGUF"},{"id":"unsloth/LFM2.5-24B-A2B","kind":"Unsloth"}],"x_mentions":[],"ollama":"https://ollama.com/library/lfm2.5","est":{"weights":{"Q4":14.4,"Q8":25.5,"FP16":47.9},"kv_gb":0.7,"kv_src":"config","need":{"Q4":16.1,"Q8":27.2,"FP16":49.6},"fit":{"gpu12":null,"gpu16":null,"gpu24":null,"gpu32":"Q8","gpu2x24":"Q8","spark1":"FP16","spark2":"FP16","mac32":"Q4","mac64":"Q8","mac128":"FP16","mac192":"FP16","mac512":"FP16","cpu":"FP16"}}},{"type":"model","id":"nvidia/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-BF16","name":"NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-BF16","org":"nvidia","date":"2026-10-12","license":"openmdw-1.1","params":124453395976,"params_src":"safetensors","active":12000000000.0,"ctx":1048576,"kv_bytes_tok":null,"dtypes":["BF16","F32","I64"],"pipeline":"image-text-to-text","modality":"vision","summary":"NVIDIA Nemotron\u2122 is a family of open models with open weights, training data, and recipes, delivering leading efficiency and accuracy for building specialized AI agents.","url":"https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-BF16","likes":30,"downloads":142,"gated":false,"quants":[{"id":"ggml-org/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-GGUF","kind":"GGUF"},{"id":"nvidia/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-NVFP4-LiLiCorr","kind":"quant"},{"id":"unsloth/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B","kind":"Unsloth"},{"id":"unsloth/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-GGUF","kind":"GGUF"},{"id":"unsloth/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-NVFP4","kind":"Unsloth"}],"x_mentions":[],"ollama":"https://ollama.com/library/nemotron-3.5-super","est":{"weights":{"Q4":74.7,"Q8":132.2,"FP16":248.9},"kv_gb":10.6,"kv_src":"rough","need":{"Q4":86.3,"Q8":143.8,"FP16":260.5},"fit":{"gpu12":null,"gpu16":null,"gpu24":null,"gpu32":null,"gpu2x24":null,"spark1":"Q4","spark2":"Q8","mac32":null,"mac64":null,"mac128":"Q4","mac192":"Q8","mac512":"FP16","cpu":null}}},{"type":"model","id":"nvidia/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-Base-BF16","name":"NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-Base-BF16","org":"nvidia","date":"2026-10-12","license":"openmdw-1.1","params":124453395974,"params_src":"safetensors","active":12000000000.0,"ctx":262144,"kv_bytes_tok":null,"dtypes":["BF16","F32"],"pipeline":"image-text-to-text","modality":"vision","summary":"NVIDIA Nemotron\u2122 is a family of open models with open weights, training data, and recipes, delivering leading efficiency and accuracy for building specialized AI agents.","url":"https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-Base-BF16","likes":16,"downloads":4,"gated":false,"quants":[],"x_mentions":[],"ollama":"https://ollama.com/library/nemotron-3.5-super","est":{"weights":{"Q4":74.7,"Q8":132.2,"FP16":248.9},"kv_gb":10.6,"kv_src":"rough","need":{"Q4":86.3,"Q8":143.8,"FP16":260.5},"fit":{"gpu12":null,"gpu16":null,"gpu24":null,"gpu32":null,"gpu2x24":null,"spark1":"Q4","spark2":"Q8","mac32":null,"mac64":null,"mac128":"Q4","mac192":"Q8","mac512":"FP16","cpu":null}}},{"type":"model","id":"allenai/HiRO-ACEv2","name":"HiRO-ACEv2","org":"allenai","date":"2026-10-12","license":"apache-2.0","params":null,"params_src":null,"active":null,"ctx":null,"kv_bytes_tok":null,"dtypes":[],"pipeline":"","modality":"","summary":"The HiRO-ACEv2 framework enables efficient generation of 3 km precipitation, 10 m eastward and northward winds, and sea level pressure fields over decades of simulated climate and arbitrary regions of the globe.","url":"https://huggingface.co/allenai/HiRO-ACEv2","likes":3,"downloads":0,"gated":false,"quants":[],"x_mentions":[],"est":null},{"type":"model","id":"autotrust/JEV-35B","name":"JEV-35B","org":"autotrust","date":"2026-10-11","license":"apache-2.0","params":35951822704,"params_src":"safetensors","active":null,"ctx":262144,"kv_bytes_tok":81920,"dtypes":["BF16","F32"],"pipeline":"text-classification","modality":"text","summary":"If you call vLLM's /v1/completions yourself instead of /v1/decide, pass topk: 0 and topp: 1.0 so that the returned probabilities are not truncated, then add the head bias and apply the temperature as below.","url":"https://huggingface.co/autotrust/JEV-35B","likes":943,"downloads":41837,"gated":false,"quants":[],"x_mentions":[],"est":{"weights":{"Q4":21.6,"Q8":38.2,"FP16":71.9},"kv_gb":0.7,"kv_src":"config","need":{"Q4":23.3,"Q8":39.9,"FP16":73.6},"fit":{"gpu12":null,"gpu16":null,"gpu24":null,"gpu32":null,"gpu2x24":"Q8","spark1":"FP16","spark2":"FP16","mac32":"Q4","mac64":"Q8","mac128":"FP16","mac192":"FP16","mac512":"FP16","cpu":"Q8"}}},{"type":"model","id":"vllm-sr/d3","name":"d3","org":"vllm-sr","date":"2026-10-10","license":"apache-2.0","params":26085330160,"params_src":"safetensors","active":null,"ctx":262144,"kv_bytes_tok":262144,"dtypes":["BF16"],"pipeline":"zero-shot-classification","modality":"other","summary":"The probabilities are this package's, as model.systemone() gives them, and confidence is the chosen option's probability; a yes / no answer's noul is the probability of yes.","url":"https://huggingface.co/vllm-sr/d3","likes":195,"downloads":2747,"gated":false,"quants":[{"id":"ggml-org/vllm-d3-GGUF","kind":"GGUF"}],"x_mentions":[],"est":{"weights":{"Q4":15.7,"Q8":27.7,"FP16":52.2},"kv_gb":2.1,"kv_src":"config","need":{"Q4":18.8,"Q8":30.8,"FP16":55.3},"fit":{"gpu12":null,"gpu16":null,"gpu24":null,"gpu32":"Q8","gpu2x24":"Q8","spark1":"FP16","spark2":"FP16","mac32":"Q4","mac64":"Q8","mac128":"FP16","mac192":"FP16","mac512":"FP16","cpu":"FP16"}}},{"type":"model","id":"datalab-to/indra-ocr","name":"indra-ocr","org":"datalab-to","date":"2026-10-10","license":"openrail","params":1236366,"params_src":"safetensors","active":null,"ctx":null,"kv_bytes_tok":null,"dtypes":["F32"],"pipeline":"image-to-text","modality":"vision","summary":"Indra OCR is a 1M-parameter OCR model that reads up to 49 pages per second on CPU.","url":"https://huggingface.co/datalab-to/indra-ocr","likes":93,"downloads":0,"gated":false,"quants":[],"x_mentions":[],"est":{"weights":{"Q4":0.0,"Q8":0.0,"FP16":0.0},"kv_gb":0.5,"kv_src":"rough","need":{"Q4":1.5,"Q8":1.5,"FP16":1.5},"fit":{"gpu12":"FP16","gpu16":"FP16","gpu24":"FP16","gpu32":"FP16","gpu2x24":"FP16","spark1":"FP16","spark2":"FP16","mac32":"FP16","mac64":"FP16","mac128":"FP16","mac192":"FP16","mac512":"FP16","cpu":"FP16"}}},{"type":"model","id":"dmis-lab/UltraBERT-380M","name":"UltraBERT-380M","org":"dmis-lab","date":"2026-10-10","license":"apache-2.0","params":429966336,"params_src":"safetensors","active":null,"ctx":null,"kv_bytes_tok":98304,"dtypes":["F32"],"pipeline":"fill-mask","modality":"other","summary":"UltraBERT is a family of English bidirectional encoders with a native 32K context that reach state of the art on both classification and retrieval, two task families on which previous encoders had to choose a side.","url":"https://huggingface.co/dmis-lab/UltraBERT-380M","likes":49,"downloads":982,"gated":false,"quants":[],"x_mentions":[],"est":{"weights":{"Q4":0.3,"Q8":0.5,"FP16":0.9},"kv_gb":0.8,"kv_src":"config","need":{"Q4":2.1,"Q8":2.3,"FP16":2.7},"fit":{"gpu12":"FP16","gpu16":"FP16","gpu24":"FP16","gpu32":"FP16","gpu2x24":"FP16","spark1":"FP16","spark2":"FP16","mac32":"FP16","mac64":"FP16","mac128":"FP16","mac192":"FP16","mac512":"FP16","cpu":"FP16"}}},{"type":"model","id":"dmis-lab/UltraBERT-150M","name":"UltraBERT-150M","org":"dmis-lab","date":"2026-10-10","license":"apache-2.0","params":189960960,"params_src":"safetensors","active":null,"ctx":null,"kv_bytes_tok":73728,"dtypes":["F32"],"pipeline":"fill-mask","modality":"other","summary":"UltraBERT is a family of English bidirectional encoders with a native 32K context that reach state of the art on both classification and retrieval, two task families on which previous encoders had to choose a side.","url":"https://huggingface.co/dmis-lab/UltraBERT-150M","likes":41,"downloads":1032,"gated":false,"quants":[],"x_mentions":[],"est":{"weights":{"Q4":0.1,"Q8":0.2,"FP16":0.4},"kv_gb":0.6,"kv_src":"config","need":{"Q4":1.7,"Q8":1.8,"FP16":2.0},"fit":{"gpu12":"FP16","gpu16":"FP16","gpu24":"FP16","gpu32":"FP16","gpu2x24":"FP16","spark1":"FP16","spark2":"FP16","mac32":"FP16","mac64":"FP16","mac128":"FP16","mac192":"FP16","mac512":"FP16","cpu":"FP16"}}},{"type":"model","id":"Qwen/Qwen-Image-2.1-Turbo","name":"Qwen-Image-2.1-Turbo","org":"Qwen","date":"2026-10-09","license":"qwen-research","params":7115124736,"params_src":"safetensors","active":null,"ctx":null,"kv_bytes_tok":null,"dtypes":["BF16"],"pipeline":"text-to-image","modality":"image","summary":"The checkpoint includes its recommended sampling schedule, so it is ready to use without manually configuring the scheduler.","url":"https://huggingface.co/Qwen/Qwen-Image-2.1-Turbo","likes":862,"downloads":15315,"gated":false,"quants":[{"id":"unsloth/Qwen-Image-2.1-Turbo-FP8","kind":"Unsloth"},{"id":"unsloth/Qwen-Image-2.1-Turbo-GGUF","kind":"GGUF"}],"x_mentions":[],"ollama":"https://ollama.com/library/qwen","est":{"weights":{"Q4":4.3,"Q8":7.6,"FP16":14.2},"kv_gb":0.6,"kv_src":"rough","need":{"Q4":5.9,"Q8":9.2,"FP16":15.8},"fit":{"gpu12":"Q8","gpu16":"Q8","gpu24":"FP16","gpu32":"FP16","gpu2x24":"FP16","spark1":"FP16","spark2":"FP16","mac32":"FP16","mac64":"FP16","mac128":"FP16","mac192":"FP16","mac512":"FP16","cpu":"FP16"}}},{"type":"model","id":"Cloudflare/clef-omni","name":"clef-omni","org":"Cloudflare","date":"2026-10-09","license":"apache-2.0","params":35259818545,"params_src":"safetensors","active":null,"ctx":null,"kv_bytes_tok":null,"dtypes":["BF16"],"pipeline":"image-text-to-text","modality":"vision","summary":"Clef-Omni is a 30B-A3B mixture-of-experts multimodal model that turns a state and a schema of typed questions into decisions.","url":"https://huggingface.co/Cloudflare/clef-omni","likes":121,"downloads":2949,"gated":false,"quants":[{"id":"mlx-community/clef-omni-4bit","kind":"MLX"},{"id":"mlx-community/clef-omni-8bit","kind":"MLX"}],"x_mentions":[],"ollama":"https://ollama.com/library/clef","est":{"weights":{"Q4":21.2,"Q8":37.5,"FP16":70.5},"kv_gb":3.0,"kv_src":"rough","need":{"Q4":25.2,"Q8":41.5,"FP16":74.5},"fit":{"gpu12":null,"gpu16":null,"gpu24":null,"gpu32":null,"gpu2x24":"Q8","spark1":"FP16","spark2":"FP16","mac32":null,"mac64":"Q8","mac128":"FP16","mac192":"FP16","mac512":"FP16","cpu":"Q8"}}},{"type":"model","id":"tencent/Youtu-Parsing-Omni","name":"Youtu-Parsing-Omni","org":"tencent","date":"2026-10-09","license":"youtu-parsing","params":5334472192,"params_src":"safetensors","active":null,"ctx":1048576,"kv_bytes_tok":46080,"dtypes":["BF16"],"pipeline":"image-text-to-text","modality":"vision","summary":"Highlights (see the technical report for details):","url":"https://huggingface.co/tencent/Youtu-Parsing-Omni","likes":99,"downloads":1329,"gated":false,"quants":[],"x_mentions":[],"est":{"weights":{"Q4":3.2,"Q8":5.7,"FP16":10.7},"kv_gb":0.4,"kv_src":"config","need":{"Q4":4.6,"Q8":7.1,"FP16":12.1},"fit":{"gpu12":"Q8","gpu16":"FP16","gpu24":"FP16","gpu32":"FP16","gpu2x24":"FP16","spark1":"FP16","spark2":"FP16","mac32":"FP16","mac64":"FP16","mac128":"FP16","mac192":"FP16","mac512":"FP16","cpu":"FP16"}}},{"type":"model","id":"LiquidAI/LFM2.5-24B-A2B-Base","name":"LFM2.5-24B-A2B-Base","org":"LiquidAI","date":"2026-10-09","license":"lfm1.0","params":23971585280,"params_src":"safetensors","active":2000000000.0,"ctx":131072,"kv_bytes_tok":81920,"dtypes":["F32","BF16"],"pipeline":"text-generation","modality":"text","summary":"LFM2.5 is a new family of hybrid models designed for on-device deployment.","url":"https://huggingface.co/LiquidAI/LFM2.5-24B-A2B-Base","likes":33,"downloads":225,"gated":false,"quants":[],"x_mentions":[],"ollama":"https://ollama.com/library/lfm2.5","est":{"weights":{"Q4":14.4,"Q8":25.5,"FP16":47.9},"kv_gb":0.7,"kv_src":"config","need":{"Q4":16.1,"Q8":27.2,"FP16":49.6},"fit":{"gpu12":null,"gpu16":null,"gpu24":null,"gpu32":"Q8","gpu2x24":"Q8","spark1":"FP16","spark2":"FP16","mac32":"Q4","mac64":"Q8","mac128":"FP16","mac192":"FP16","mac512":"FP16","cpu":"FP16"}}},{"type":"model","id":"nvidia/NVIDIA-Nemotron-3.5-Bolt-74B-A9B-BF16","name":"NVIDIA-Nemotron-3.5-Bolt-74B-A9B-BF16","org":"nvidia","date":"2026-10-09","license":"openmdw-1.1","params":74885081096,"params_src":"safetensors","active":9000000000.0,"ctx":262144,"kv_bytes_tok":null,"dtypes":["F32","BF16","I64"],"pipeline":"image-text-to-text","modality":"vision","summary":"NVIDIA Nemotron 3.5 Bolt 74B-A9B is a vision-language model (VLM) designed to run everyday agentic tasks locally.","url":"https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Bolt-74B-A9B-BF16","likes":18,"downloads":4,"gated":false,"quants":[{"id":"nvidia/NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB-GGUF","kind":"GGUF"},{"id":"unsloth/NVIDIA-Nemotron-3.5-Bolt-74B-A9B-GGUF","kind":"GGUF"}],"x_mentions":[],"ollama":"https://ollama.com/library/nemotron-3.5-bolt","est":{"weights":{"Q4":44.9,"Q8":79.6,"FP16":149.8},"kv_gb":6.4,"kv_src":"rough","need":{"Q4":52.3,"Q8":87.0,"FP16":157.2},"fit":{"gpu12":null,"gpu16":null,"gpu24":null,"gpu32":null,"gpu2x24":null,"spark1":"Q8","spark2":"FP16","mac32":null,"mac64":null,"mac128":"Q8","mac192":"Q8","mac512":"FP16","cpu":"Q4"}}},{"type":"model","id":"cloud0day3/antalia-mini","name":"antalia-mini","org":"cloud0day3","date":"2026-10-08","license":"apache-2.0","params":null,"params_src":null,"active":null,"ctx":null,"kv_bytes_tok":null,"dtypes":[],"pipeline":"text-to-speech","modality":"audio","summary":"Antalia-2 Mini (version 1.0.1) is a small Turkish text-to-speech model with one voice.","url":"https://huggingface.co/cloud0day3/antalia-mini","likes":43,"downloads":1996,"gated":false,"quants":[],"x_mentions":[],"est":null},{"type":"model","id":"mistralai/Voxtral-Mini-4B-Realtime-Arabic","name":"Voxtral-Mini-4B-Realtime-Arabic","org":"mistralai","date":"2026-10-08","license":"","params":4429679360,"params_src":"safetensors","active":null,"ctx":131072,"kv_bytes_tok":106496,"dtypes":["BF16"],"pipeline":"automatic-speech-recognition","modality":"audio","summary":"Voxtral Mini 4B Realtime Arabic is a streaming speech-to-text model for Arabic dialects and Modern Standard Arabic.","url":"https://huggingface.co/mistralai/Voxtral-Mini-4B-Realtime-Arabic","likes":33,"downloads":829,"gated":false,"quants":[],"x_mentions":[],"est":{"weights":{"Q4":2.7,"Q8":4.7,"FP16":8.9},"kv_gb":0.9,"kv_src":"config","need":{"Q4":4.6,"Q8":6.6,"FP16":10.8},"fit":{"gpu12":"FP16","gpu16":"FP16","gpu24":"FP16","gpu32":"FP16","gpu2x24":"FP16","spark1":"FP16","spark2":"FP16","mac32":"FP16","mac64":"FP16","mac128":"FP16","mac192":"FP16","mac512":"FP16","cpu":"FP16"}}},{"type":"model","id":"mistralai/LIDstral-Arabic","name":"LIDstral-Arabic","org":"mistralai","date":"2026-10-08","license":"apache-2.0","params":null,"params_src":null,"active":null,"ctx":null,"kv_bytes_tok":null,"dtypes":[],"pipeline":"text-classification","modality":"text","summary":"LIDstral Arabic is a fast language and dialect classifier for text written in Arabic script.","url":"https://huggingface.co/mistralai/LIDstral-Arabic","likes":20,"downloads":85,"gated":false,"quants":[],"x_mentions":[],"est":null},{"type":"model","id":"openbmb/MiniCPM-V-4.7-35B-A3B","name":"MiniCPM-V-4.7-35B-A3B","org":"openbmb","date":"2026-10-07","license":"apache-2.0","params":35212875824,"params_src":"safetensors","active":3000000000.0,"ctx":262144,"kv_bytes_tok":81920,"dtypes":["BF16"],"pipeline":"image-text-to-text","modality":"vision","summary":"An MoE model can have far more total parameters than it activates for a single token, but its on-device storage requirements are not limited to those active parameters.","url":"https://huggingface.co/openbmb/MiniCPM-V-4.7-35B-A3B","likes":782,"downloads":9278,"gated":false,"quants":[{"id":"mlx-community/MiniCPM-V-4.7-35B-A3B-4bit","kind":"MLX"},{"id":"mlx-community/MiniCPM-V-4.7-35B-A3B-5bit","kind":"MLX"},{"id":"mlx-community/MiniCPM-V-4.7-35B-A3B-6bit","kind":"MLX"},{"id":"mlx-community/MiniCPM-V-4.7-35B-A3B-8bit","kind":"MLX"},{"id":"mlx-community/MiniCPM-V-4.7-35B-A3B-OptiQ-4bit","kind":"MLX"},{"id":"mlx-community/MiniCPM-V-4.7-35B-A3B-mxfp4","kind":"MLX"},{"id":"mlx-community/MiniCPM-V-4.7-35B-A3B-mxfp8","kind":"MLX"},{"id":"mlx-community/MiniCPM-V-4.7-35B-A3B-nvfp4","kind":"MLX"},{"id":"openbmb/MiniCPM-V-4.7-35B-A3B-gguf","kind":"GGUF"}],"x_mentions":[],"ollama":"https://ollama.com/library/minicpm-v","est":{"weights":{"Q4":21.1,"Q8":37.4,"FP16":70.4},"kv_gb":0.7,"kv_src":"config","need":{"Q4":22.8,"Q8":39.1,"FP16":72.1},"fit":{"gpu12":null,"gpu16":null,"gpu24":null,"gpu32":null,"gpu2x24":"Q8","spark1":"FP16","spark2":"FP16","mac32":"Q4","mac64":"Q8","mac128":"FP16","mac192":"FP16","mac512":"FP16","cpu":"Q8"}}},{"type":"model","id":"lightonai/LightOnOCR-3-0.8B","name":"LightOnOCR-3-0.8B","org":"lightonai","date":"2026-10-07","license":"apache-2.0","params":852985920,"params_src":"safetensors","active":null,"ctx":262144,"kv_bytes_tok":49152,"dtypes":["BF16"],"pipeline":"image-text-to-text","modality":"vision","summary":"LightOnOCR-3 is a new family of highly performant lightweight OCR models.","url":"https://huggingface.co/lightonai/LightOnOCR-3-0.8B","likes":65,"downloads":3463,"gated":false,"quants":[],"x_mentions":[],"est":{"weights":{"Q4":0.5,"Q8":0.9,"FP16":1.7},"kv_gb":0.4,"kv_src":"config","need":{"Q4":1.9,"Q8":2.3,"FP16":3.1},"fit":{"gpu12":"FP16","gpu16":"FP16","gpu24":"FP16","gpu32":"FP16","gpu2x24":"FP16","spark1":"FP16","spark2":"FP16","mac32":"FP16","mac64":"FP16","mac128":"FP16","mac192":"FP16","mac512":"FP16","cpu":"FP16"}}},{"type":"model","id":"openbmb/MiniCPM-V-4.7-1B","name":"MiniCPM-V-4.7-1B","org":"openbmb","date":"2026-10-06","license":"apache-2.0","params":1554476272,"params_src":"safetensors","active":null,"ctx":262144,"kv_bytes_tok":49152,"dtypes":["BF16"],"pipeline":"image-text-to-text","modality":"vision","summary":"The 1B variant of MiniCPM-V 4.7 advances the performance\u2013efficiency Pareto frontier.","url":"https://huggingface.co/openbmb/MiniCPM-V-4.7-1B","likes":358,"downloads":5274,"gated":false,"quants":[{"id":"mlx-community/MiniCPM-V-4.7-1B-4bit","kind":"MLX"},{"id":"mlx-community/MiniCPM-V-4.7-1B-5bit","kind":"MLX"},{"id":"mlx-community/MiniCPM-V-4.7-1B-6bit","kind":"MLX"},{"id":"mlx-community/MiniCPM-V-4.7-1B-8bit","kind":"MLX"},{"id":"mlx-community/MiniCPM-V-4.7-1B-OptiQ-4bit","kind":"MLX"},{"id":"mlx-community/MiniCPM-V-4.7-1B-mxfp4","kind":"MLX"},{"id":"mlx-community/MiniCPM-V-4.7-1B-mxfp8","kind":"MLX"},{"id":"mlx-community/MiniCPM-V-4.7-1B-nvfp4","kind":"MLX"},{"id":"openbmb/MiniCPM-V-4.7-1B-gguf","kind":"GGUF"}],"x_mentions":[],"ollama":"https://ollama.com/library/minicpm-v","est":{"weights":{"Q4":0.9,"Q8":1.7,"FP16":3.1},"kv_gb":0.4,"kv_src":"config","need":{"Q4":2.3,"Q8":3.1,"FP16":4.5},"fit":{"gpu12":"FP16","gpu16":"FP16","gpu24":"FP16","gpu32":"FP16","gpu2x24":"FP16","spark1":"FP16","spark2":"FP16","mac32":"FP16","mac64":"FP16","mac128":"FP16","mac192":"FP16","mac512":"FP16","cpu":"FP16"}}},{"type":"model","id":"lightonai/LightOnOCR-3-4B","name":"LightOnOCR-3-4B","org":"lightonai","date":"2026-10-06","license":"apache-2.0","params":4539265536,"params_src":"safetensors","active":null,"ctx":262144,"kv_bytes_tok":131072,"dtypes":["BF16"],"pipeline":"image-text-to-text","modality":"vision","summary":"LightOnOCR-3 is a new family of highly performant lightweight OCR models.","url":"https://huggingface.co/lightonai/LightOnOCR-3-4B","likes":94,"downloads":9012,"gated":false,"quants":[],"x_mentions":[],"est":{"weights":{"Q4":2.7,"Q8":4.8,"FP16":9.1},"kv_gb":1.1,"kv_src":"config","need":{"Q4":4.8,"Q8":6.9,"FP16":11.2},"fit":{"gpu12":"Q8","gpu16":"FP16","gpu24":"FP16","gpu32":"FP16","gpu2x24":"FP16","spark1":"FP16","spark2":"FP16","mac32":"FP16","mac64":"FP16","mac128":"FP16","mac192":"FP16","mac512":"FP16","cpu":"FP16"}}},{"type":"model","id":"neuphonic/neudecide","name":"neudecide","org":"neuphonic","date":"2026-10-06","license":"apache-2.0","params":null,"params_src":null,"active":null,"ctx":null,"kv_bytes_tok":null,"dtypes":[],"pipeline":"audio-classification","modality":"other","summary":"NeuDecide is a pre-trained streaming English speech encoder joined to a compact tool-calling encoder\u2013decoder, initialised from Needle, and trained end to end as one system.","url":"https://huggingface.co/neuphonic/neudecide","likes":76,"downloads":580,"gated":true,"quants":[],"x_mentions":[],"est":null},{"type":"model","id":"LiquidAI/d1-3B","name":"d1-3B","org":"LiquidAI","date":"2026-10-05","license":"lfm1.0","params":3123483888,"params_src":"safetensors","active":null,"ctx":32768,"kv_bytes_tok":61440,"dtypes":["BF16"],"pipeline":"image-text-to-text","modality":"vision","summary":"d1-3B is a 3B parameter decision model built on LFM2.5-VL-3B.","url":"https://huggingface.co/LiquidAI/d1-3B","likes":416,"downloads":16731,"gated":false,"quants":[{"id":"LiquidAI/d1-3B-GGUF","kind":"GGUF"},{"id":"LiquidAI/d1-3B-MLX-4bit","kind":"MLX"},{"id":"LiquidAI/d1-3B-MLX-8bit","kind":"MLX"},{"id":"LiquidAI/d1-3B-MLX-fp16","kind":"MLX"}],"x_mentions":[],"ollama":"https://ollama.com/library/d1","est":{"weights":{"Q4":1.9,"Q8":3.3,"FP16":6.2},"kv_gb":0.5,"kv_src":"config","need":{"Q4":3.4,"Q8":4.8,"FP16":7.7},"fit":{"gpu12":"FP16","gpu16":"FP16","gpu24":"FP16","gpu32":"FP16","gpu2x24":"FP16","spark1":"FP16","spark2":"FP16","mac32":"FP16","mac64":"FP16","mac128":"FP16","mac192":"FP16","mac512":"FP16","cpu":"FP16"}}},{"type":"model","id":"speridlabs/iris-3b","name":"iris-3b","org":"speridlabs","date":"2026-10-05","license":"apache-2.0","params":2987511888,"params_src":"safetensors","active":null,"ctx":null,"kv_bytes_tok":null,"dtypes":["F32"],"pipeline":"text-to-image","modality":"image","summary":"Generative priors are a promising foundation for downstream vision tasks.","url":"https://huggingface.co/speridlabs/iris-3b","likes":282,"downloads":1257,"gated":false,"quants":[],"x_mentions":[],"est":{"weights":{"Q4":1.8,"Q8":3.2,"FP16":6.0},"kv_gb":0.5,"kv_src":"rough","need":{"Q4":3.3,"Q8":4.7,"FP16":7.5},"fit":{"gpu12":"FP16","gpu16":"FP16","gpu24":"FP16","gpu32":"FP16","gpu2x24":"FP16","spark1":"FP16","spark2":"FP16","mac32":"FP16","mac64":"FP16","mac128":"FP16","mac192":"FP16","mac512":"FP16","cpu":"FP16"}}},{"type":"model","id":"perplexity-ai/pplx-decider-v1.1-27b","name":"pplx-decider-v1.1-27b","org":"perplexity-ai","date":"2026-10-05","license":"apache-2.0","params":26085330160,"params_src":"safetensors","active":null,"ctx":262144,"kv_bytes_tok":262144,"dtypes":["BF16"],"pipeline":"text-classification","modality":"text","summary":"pplx-decider-v1.1-27b is an update of our pplx-decider-v1-27b model, that boost the Decision Index score from 56.4 to 61.56, now outperforming Jev by more than 3.5 points, while using the same backbone.","url":"https://huggingface.co/perplexity-ai/pplx-decider-v1.1-27b","likes":149,"downloads":3520,"gated":false,"quants":[{"id":"mlx-community/pplx-decider-v1.1-27b-4bit","kind":"MLX"}],"x_mentions":[],"ollama":"https://ollama.com/library/pplx-decider","est":{"weights":{"Q4":15.7,"Q8":27.7,"FP16":52.2},"kv_gb":2.1,"kv_src":"config","need":{"Q4":18.8,"Q8":30.8,"FP16":55.3},"fit":{"gpu12":null,"gpu16":null,"gpu24":null,"gpu32":"Q8","gpu2x24":"Q8","spark1":"FP16","spark2":"FP16","mac32":"Q4","mac64":"Q8","mac128":"FP16","mac192":"FP16","mac512":"FP16","cpu":"FP16"}}},{"type":"model","id":"LiquidAI/d1-omni-600M","name":"d1-omni-600M","org":"LiquidAI","date":"2026-10-05","license":"lfm1.0","params":587161089,"params_src":"safetensors","active":null,"ctx":128000,"kv_bytes_tok":32768,"dtypes":["F32"],"pipeline":"image-text-to-text","modality":"vision","summary":"d1-omni-600M is a 600M parameter decision model built on LFM2.5-Encoder-350M.","url":"https://huggingface.co/LiquidAI/d1-omni-600M","likes":124,"downloads":12851,"gated":false,"quants":[{"id":"LiquidAI/d1-omni-600M-GGUF","kind":"GGUF"},{"id":"LiquidAI/d1-omni-600M-MLX-8bit","kind":"MLX"},{"id":"LiquidAI/d1-omni-600M-MLX-fp16","kind":"MLX"}],"x_mentions":[],"ollama":"https://ollama.com/library/d1","est":{"weights":{"Q4":0.4,"Q8":0.6,"FP16":1.2},"kv_gb":0.3,"kv_src":"config","need":{"Q4":1.7,"Q8":1.9,"FP16":2.5},"fit":{"gpu12":"FP16","gpu16":"FP16","gpu24":"FP16","gpu32":"FP16","gpu2x24":"FP16","spark1":"FP16","spark2":"FP16","mac32":"FP16","mac64":"FP16","mac128":"FP16","mac192":"FP16","mac512":"FP16","cpu":"FP16"}}},{"type":"model","id":"inclusionAI/LemonDecider","name":"LemonDecider","org":"inclusionAI","date":"2026-10-05","license":"apache-2.0","params":4539918336,"params_src":"safetensors","active":null,"ctx":262144,"kv_bytes_tok":131072,"dtypes":["BF16"],"pipeline":"feature-extraction","modality":"embedding","summary":"A typed decision engine derived from Qwen/Qwen3.5-4B: it answers structured choice / noul / score questions in a single forward pass with an explicit unknown (abstention) candidate.","url":"https://huggingface.co/inclusionAI/LemonDecider","likes":23,"downloads":156,"gated":false,"quants":[],"x_mentions":[],"est":{"weights":{"Q4":2.7,"Q8":4.8,"FP16":9.1},"kv_gb":1.1,"kv_src":"config","need":{"Q4":4.8,"Q8":6.9,"FP16":11.2},"fit":{"gpu12":"Q8","gpu16":"FP16","gpu24":"FP16","gpu32":"FP16","gpu2x24":"FP16","spark1":"FP16","spark2":"FP16","mac32":"FP16","mac64":"FP16","mac128":"FP16","mac192":"FP16","mac512":"FP16","cpu":"FP16"}}},{"type":"model","id":"fishaudio/drama-3","name":"drama-3","org":"fishaudio","date":"2026-10-01","license":"fish-audio-research-license","params":35899643008,"params_src":"safetensors","active":null,"ctx":262144,"kv_bytes_tok":81920,"dtypes":["BF16","F8_E4M3"],"pipeline":"text-to-speech","modality":"audio","summary":"Drama 3 runs on SGLang-Omni from launch day.","url":"https://huggingface.co/fishaudio/drama-3","likes":432,"downloads":4048,"gated":false,"quants":[{"id":"mlx-community/fish-audio-drama-3-3bit","kind":"MLX"},{"id":"mlx-community/fish-audio-drama-3-4bit","kind":"MLX"},{"id":"mlx-community/fish-audio-drama-3-8bit","kind":"MLX"}],"x_mentions":[],"est":{"weights":{"Q4":21.5,"Q8":38.1,"FP16":71.8},"kv_gb":0.7,"kv_src":"config","need":{"Q4":23.2,"Q8":39.8,"FP16":73.5},"fit":{"gpu12":null,"gpu16":null,"gpu24":null,"gpu32":null,"gpu2x24":"Q8","spark1":"FP16","spark2":"FP16","mac32":"Q4","mac64":"Q8","mac128":"FP16","mac192":"FP16","mac512":"FP16","cpu":"Q8"}}},{"type":"model","id":"canberkkkkkk/ema-lightning","name":"ema-lightning","org":"canberkkkkkk","date":"2026-10-01","license":"apache-2.0","params":null,"params_src":null,"active":null,"ctx":null,"kv_bytes_tok":null,"dtypes":[],"pipeline":"text-to-speech","modality":"audio","summary":"The first chunk is one second of audio and is ready in about 4 ms on a GPU.","url":"https://huggingface.co/canberkkkkkk/ema-lightning","likes":414,"downloads":27395,"gated":false,"quants":[],"x_mentions":[],"est":null},{"type":"model","id":"nvidia/PixelUMM","name":"PixelUMM","org":"nvidia","date":"2026-10-01","license":"nvidia-one-way-noncommercial-license","params":null,"params_src":null,"active":null,"ctx":null,"kv_bytes_tok":null,"dtypes":[],"pipeline":"any-to-any","modality":"omni","summary":"PixelUMM is an NVIDIA-developed, encoder-free unified multimodal model for joint understanding and generation across text, images, and video directly in pixel space.","url":"https://huggingface.co/nvidia/PixelUMM","likes":88,"downloads":0,"gated":false,"quants":[],"x_mentions":[],"est":null},{"type":"model","id":"nvidia/PixelDiT2-ImageNet","name":"PixelDiT2-ImageNet","org":"nvidia","date":"2026-10-01","license":"","params":null,"params_src":null,"active":null,"ctx":null,"kv_bytes_tok":null,"dtypes":[],"pipeline":"","modality":"","summary":"","url":"https://huggingface.co/nvidia/PixelDiT2-ImageNet","likes":11,"downloads":0,"gated":false,"quants":[],"x_mentions":[],"est":null},{"type":"model","id":"ibm-granite/ensemble-forecasting-with-ibm-granite-time-series","name":"ensemble-forecasting-with-ibm-granite-time-series","org":"ibm-granite","date":"2026-10-01","license":"","params":null,"params_src":null,"active":null,"ctx":null,"kv_bytes_tok":null,"dtypes":[],"pipeline":"","modality":"","summary":"","url":"https://huggingface.co/ibm-granite/ensemble-forecasting-with-ibm-granite-time-series","likes":2,"downloads":0,"gated":false,"quants":[],"x_mentions":[],"est":null},{"type":"model","id":"Cloudflare/clef","name":"clef","org":"Cloudflare","date":"2026-09-30","license":"apache-2.0","params":27356728560,"params_src":"safetensors","active":null,"ctx":262144,"kv_bytes_tok":262144,"dtypes":["BF16"],"pipeline":"image-text-to-text","modality":"vision","summary":"Clef is a 27B multimodal model that turns a state and a schema of typed questions into decisions.","url":"https://huggingface.co/Cloudflare/clef","likes":2176,"downloads":19314,"gated":false,"quants":[{"id":"bartowski/Cloudflare_clef-GGUF","kind":"GGUF"},{"id":"ggml-org/Clef-GGUF","kind":"GGUF"},{"id":"mlx-community/clef-4bit","kind":"MLX"},{"id":"mlx-community/clef-8bit","kind":"MLX"},{"id":"mlx-community/clef-OptiQ-4bit","kind":"MLX"}],"x_mentions":[],"ollama":"https://ollama.com/library/clef","est":{"weights":{"Q4":16.4,"Q8":29.1,"FP16":54.7},"kv_gb":2.1,"kv_src":"config","need":{"Q4":19.5,"Q8":32.2,"FP16":57.8},"fit":{"gpu12":null,"gpu16":null,"gpu24":null,"gpu32":null,"gpu2x24":"Q8","spark1":"FP16","spark2":"FP16","mac32":"Q4","mac64":"Q8","mac128":"FP16","mac192":"FP16","mac512":"FP16","cpu":"Q8"}}},{"type":"model","id":"Cloudflare/clef-flash","name":"clef-flash","org":"Cloudflare","date":"2026-09-30","license":"apache-2.0","params":9409813744,"params_src":"safetensors","active":null,"ctx":262144,"kv_bytes_tok":131072,"dtypes":["BF16"],"pipeline":"image-text-to-text","modality":"vision","summary":"Clef-Flash is a 9B multimodal model that turns a state and a schema of typed questions into decisions.","url":"https://huggingface.co/Cloudflare/clef-flash","likes":824,"downloads":28485,"gated":false,"quants":[{"id":"bartowski/Cloudflare_clef-flash-GGUF","kind":"GGUF"},{"id":"ggml-org/Clef-Flash-GGUF","kind":"GGUF"},{"id":"mlx-community/clef-flash-4bit","kind":"MLX"},{"id":"mlx-community/clef-flash-8bit","kind":"MLX"},{"id":"mlx-community/clef-flash-OptiQ-4bit","kind":"MLX"}],"x_mentions":[],"ollama":"https://ollama.com/library/clef-flash","est":{"weights":{"Q4":5.6,"Q8":10.0,"FP16":18.8},"kv_gb":1.1,"kv_src":"config","need":{"Q4":7.7,"Q8":12.1,"FP16":20.9},"fit":{"gpu12":"Q4","gpu16":"Q8","gpu24":"FP16","gpu32":"FP16","gpu2x24":"FP16","spark1":"FP16","spark2":"FP16","mac32":"FP16","mac64":"FP16","mac128":"FP16","mac192":"FP16","mac512":"FP16","cpu":"FP16"}}},{"type":"model","id":"Cactus-Compute/whistle","name":"whistle","org":"Cactus-Compute","date":"2026-09-30","license":"apache-2.0","params":null,"params_src":null,"active":null,"ctx":null,"kv_bytes_tok":null,"dtypes":[],"pipeline":"automatic-speech-recognition","modality":"audio","summary":"A speech-to-text model for mobiles, wearables, robots, smart home, automotive and microcontrollers.","url":"https://huggingface.co/Cactus-Compute/whistle","likes":413,"downloads":38794,"gated":false,"quants":[],"x_mentions":[],"est":null},{"type":"model","id":"KittenML/kitten-tts-2","name":"kitten-tts-2","org":"KittenML","date":"2026-09-30","license":"stellon-labs-community-license","params":null,"params_src":null,"active":null,"ctx":null,"kv_bytes_tok":null,"dtypes":[],"pipeline":"text-to-speech","modality":"audio","summary":"Only these twenty tags are recognised.","url":"https://huggingface.co/KittenML/kitten-tts-2","likes":92,"downloads":25472,"gated":false,"quants":[],"x_mentions":[],"est":null},{"type":"model","id":"microsoft/AesCode-8B","name":"AesCode-8B","org":"microsoft","date":"2026-09-29","license":"apache-2.0","params":8767123696,"params_src":"safetensors","active":null,"ctx":262144,"kv_bytes_tok":147456,"dtypes":["BF16"],"pipeline":"image-text-to-text","modality":"vision","summary":"We evaluate on 300 infographic samples.","url":"https://huggingface.co/microsoft/AesCode-8B","likes":63,"downloads":476,"gated":false,"quants":[{"id":"bartowski/AesCode-8B-GGUF","kind":"GGUF"}],"x_mentions":[],"est":{"weights":{"Q4":5.3,"Q8":9.3,"FP16":17.5},"kv_gb":1.2,"kv_src":"config","need":{"Q4":7.5,"Q8":11.5,"FP16":19.7},"fit":{"gpu12":"Q4","gpu16":"Q8","gpu24":"FP16","gpu32":"FP16","gpu2x24":"FP16","spark1":"FP16","spark2":"FP16","mac32":"FP16","mac64":"FP16","mac128":"FP16","mac192":"FP16","mac512":"FP16","cpu":"FP16"}}},{"type":"model","id":"microsoft/AesCode-32B","name":"AesCode-32B","org":"microsoft","date":"2026-09-29","license":"apache-2.0","params":33357390064,"params_src":"safetensors","active":null,"ctx":262144,"kv_bytes_tok":262144,"dtypes":["BF16"],"pipeline":"image-text-to-text","modality":"vision","summary":"AesCode-32B starts from Qwen3-VL-32B-Instruct and is trained with cold-start SFT followed by GDPO across seven reward channels.","url":"https://huggingface.co/microsoft/AesCode-32B","likes":49,"downloads":503,"gated":false,"quants":[{"id":"bartowski/AesCode-32B-GGUF","kind":"GGUF"}],"x_mentions":[],"est":{"weights":{"Q4":20.0,"Q8":35.4,"FP16":66.7},"kv_gb":2.1,"kv_src":"config","need":{"Q4":23.1,"Q8":38.5,"FP16":69.8},"fit":{"gpu12":null,"gpu16":null,"gpu24":null,"gpu32":null,"gpu2x24":"Q8","spark1":"FP16","spark2":"FP16","mac32":"Q4","mac64":"Q8","mac128":"FP16","mac192":"FP16","mac512":"FP16","cpu":"Q8"}}},{"type":"model","id":"XiaomiMiMo/MiMo-V2.6-Flash-MOPD","name":"MiMo-V2.6-Flash-MOPD","org":"XiaomiMiMo","date":"2026-09-27","license":"mit","params":310756322688,"params_src":"safetensors","active":null,"ctx":1048576,"kv_bytes_tok":147456,"dtypes":["F32","BF16","F8_E4M3","U8"],"pipeline":"text-generation","modality":"text","summary":"MOPD2 distills several domain-specialized teachers into the student on-policy.","url":"https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Flash-MOPD","likes":65,"downloads":8741,"gated":false,"quants":[{"id":"ggml-org/MiMo-V2.6-Flash-MOPD-GGUF","kind":"GGUF"}],"x_mentions":[],"est":{"weights":{"Q4":186.5,"Q8":330.2,"FP16":621.5},"kv_gb":1.2,"kv_src":"config","need":{"Q4":188.7,"Q8":332.4,"FP16":623.7},"fit":{"gpu12":null,"gpu16":null,"gpu24":null,"gpu32":null,"gpu2x24":null,"spark1":null,"spark2":"Q4","mac32":null,"mac64":null,"mac128":null,"mac192":null,"mac512":"Q8","cpu":null}}},{"type":"model","id":"XiaomiMiMo/MiMo-V2.6-Pro-MOPD","name":"MiMo-V2.6-Pro-MOPD","org":"XiaomiMiMo","date":"2026-09-27","license":"mit","params":1024216603392,"params_src":"safetensors","active":null,"ctx":1048576,"kv_bytes_tok":430080,"dtypes":["BF16","F32","F8_E4M3","U8"],"pipeline":"text-generation","modality":"text","summary":"MOPD2 distills several domain-specialized teachers into the student on-policy.","url":"https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Pro-MOPD","likes":34,"downloads":5938,"gated":false,"quants":[{"id":"ggml-org/MiMo-V2.6-Pro-MOPD-GGUF","kind":"GGUF"}],"x_mentions":[],"est":{"weights":{"Q4":614.5,"Q8":1088.2,"FP16":2048.4},"kv_gb":3.5,"kv_src":"config","need":{"Q4":619.0,"Q8":1092.7,"FP16":2052.9},"fit":{"gpu12":null,"gpu16":null,"gpu24":null,"gpu32":null,"gpu2x24":null,"spark1":null,"spark2":null,"mac32":null,"mac64":null,"mac128":null,"mac192":null,"mac512":null,"cpu":null}}},{"type":"model","id":"FrancisRing/Prism","name":"Prism","org":"FrancisRing","date":"2026-09-24","license":"mit","params":null,"params_src":null,"active":null,"ctx":null,"kv_bytes_tok":null,"dtypes":[],"pipeline":"image-to-video","modality":"video","summary":"Regarding the bucketed CSV files from step2 in \ud83e\uddf1Training Dataset Preparation, they are the sole data manifest consumed by the trainer.","url":"https://huggingface.co/FrancisRing/Prism","likes":177,"downloads":0,"gated":false,"quants":[],"x_mentions":[],"est":null},{"type":"model","id":"madebyollin/texture-fix-vae-for-qwen-image-2.1","name":"texture-fix-vae-for-qwen-image-2.1","org":"madebyollin","date":"2026-09-24","license":"qwen-research","params":337740404,"params_src":"safetensors","active":null,"ctx":null,"kv_bytes_tok":null,"dtypes":["F32"],"pipeline":"","modality":"","summary":"Texture-Fix-VAE-for-Qwen-Image-2.1 is the Qwen-Image-2.1 VAE, but finetuned to produce cleaner textures with no checkerboard artifacts (and no NaNs when running in fp16).","url":"https://huggingface.co/madebyollin/texture-fix-vae-for-qwen-image-2.1","likes":175,"downloads":31492,"gated":false,"quants":[],"x_mentions":[],"ollama":"https://ollama.com/library/qwen","est":{"weights":{"Q4":0.2,"Q8":0.4,"FP16":0.7},"kv_gb":0.5,"kv_src":"rough","need":{"Q4":1.7,"Q8":1.9,"FP16":2.2},"fit":{"gpu12":"FP16","gpu16":"FP16","gpu24":"FP16","gpu32":"FP16","gpu2x24":"FP16","spark1":"FP16","spark2":"FP16","mac32":"FP16","mac64":"FP16","mac128":"FP16","mac192":"FP16","mac512":"FP16","cpu":"FP16"}}},{"type":"model","id":"inclusionAI/AI-Transparency","name":"AI-Transparency","org":"inclusionAI","date":"2026-09-24","license":"","params":null,"params_src":null,"active":null,"ctx":null,"kv_bytes_tok":null,"dtypes":[],"pipeline":"","modality":"","summary":"This repository provides public training-content summaries for the model versions identified in the documents below.","url":"https://huggingface.co/inclusionAI/AI-Transparency","likes":10,"downloads":0,"gated":false,"quants":[],"x_mentions":[],"est":null},{"type":"model","id":"AxiomicLabs/GPT-X3-150M","name":"GPT-X3-150M","org":"AxiomicLabs","date":"2026-09-23","license":"apache-2.0","params":149400930,"params_src":"safetensors","active":null,"ctx":8192,"kv_bytes_tok":18432,"dtypes":["F32"],"pipeline":"text-generation","modality":"text","summary":"GPT-X3-150M is Axiomic Labs\u2019 third-generation base model, built on the TX5 hybrid GDN2/gated GQA architecture and trained from scratch with TrainWork.","url":"https://huggingface.co/AxiomicLabs/GPT-X3-150M","likes":65,"downloads":993,"gated":false,"quants":[],"x_mentions":[],"est":{"weights":{"Q4":0.1,"Q8":0.2,"FP16":0.3},"kv_gb":0.2,"kv_src":"config","need":{"Q4":1.3,"Q8":1.4,"FP16":1.5},"fit":{"gpu12":"FP16","gpu16":"FP16","gpu24":"FP16","gpu32":"FP16","gpu2x24":"FP16","spark1":"FP16","spark2":"FP16","mac32":"FP16","mac64":"FP16","mac128":"FP16","mac192":"FP16","mac512":"FP16","cpu":"FP16"}}},{"type":"model","id":"ibm-granite/granite-timeseries-ensemble-r1","name":"granite-timeseries-ensemble-r1","org":"ibm-granite","date":"2026-09-23","license":"apache-2.0","params":null,"params_src":null,"active":null,"ctx":null,"kv_bytes_tok":null,"dtypes":[],"pipeline":"time-series-forecasting","modality":"other","summary":"Granite Timeseries Ensemble R1 is an extensible framework for combining quantile forecasts from multiple pretrained time-series models.","url":"https://huggingface.co/ibm-granite/granite-timeseries-ensemble-r1","likes":5,"downloads":138,"gated":false,"quants":[],"x_mentions":[],"est":null},{"type":"model","id":"swiss-ai/Apertus-v1.5-70B-base","name":"Apertus-v1.5-70B-base","org":"swiss-ai","date":"2026-09-23","license":"apache-2.0","params":71942010529,"params_src":"safetensors","active":null,"ctx":262144,"kv_bytes_tok":327680,"dtypes":["BF16","F32"],"pipeline":"image-text-to-text","modality":"vision","summary":"Apertus 1.5 is a family of 8B and 70B fully open, multilingual, and multimodal models from the Swiss AI Initiative, trained only on openly available data and released with open weights, open data, and open training documentation.","url":"https://huggingface.co/swiss-ai/Apertus-v1.5-70B-base","likes":3,"downloads":414,"gated":false,"quants":[],"x_mentions":[],"est":{"weights":{"Q4":43.2,"Q8":76.4,"FP16":143.9},"kv_gb":2.7,"kv_src":"config","need":{"Q4":46.9,"Q8":80.1,"FP16":147.6},"fit":{"gpu12":null,"gpu16":null,"gpu24":null,"gpu32":null,"gpu2x24":null,"spark1":"Q8","spark2":"FP16","mac32":null,"mac64":"Q4","mac128":"Q8","mac192":"Q8","mac512":"FP16","cpu":"Q4"}}},{"type":"model","id":"swiss-ai/Apertus-v1.5-8B-base","name":"Apertus-v1.5-8B-base","org":"swiss-ai","date":"2026-09-23","license":"apache-2.0","params":8839738945,"params_src":"safetensors","active":null,"ctx":262144,"kv_bytes_tok":131072,"dtypes":["BF16","F32"],"pipeline":"image-text-to-text","modality":"vision","summary":"Apertus 1.5 is a family of 8B and 70B fully open, multilingual, and multimodal models from the Swiss AI Initiative, trained only on openly available data and released with open weights, open data, and open training documentation.","url":"https://huggingface.co/swiss-ai/Apertus-v1.5-8B-base","likes":3,"downloads":1319,"gated":false,"quants":[],"x_mentions":[],"est":{"weights":{"Q4":5.3,"Q8":9.4,"FP16":17.7},"kv_gb":1.1,"kv_src":"config","need":{"Q4":7.4,"Q8":11.5,"FP16":19.8},"fit":{"gpu12":"Q4","gpu16":"Q8","gpu24":"FP16","gpu32":"FP16","gpu2x24":"FP16","spark1":"FP16","spark2":"FP16","mac32":"FP16","mac64":"FP16","mac128":"FP16","mac192":"FP16","mac512":"FP16","cpu":"FP16"}}},{"type":"model","id":"XiaomiMiMo/MiMo-V2.6-Pro-RL","name":"MiMo-V2.6-Pro-RL","org":"XiaomiMiMo","date":"2026-09-21","license":"mit","params":1024216603392,"params_src":"safetensors","active":null,"ctx":1048576,"kv_bytes_tok":430080,"dtypes":["BF16","F32","F8_E4M3","U8"],"pipeline":"text-generation","modality":"text","summary":"MiMo-V2.6-Pro-RL is the flagship checkpoint of the MiMo-V2.6 series.","url":"https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Pro-RL","likes":729,"downloads":99548,"gated":false,"quants":[{"id":"ggml-org/MiMo-V2.6-Pro-RL-GGUF","kind":"GGUF"},{"id":"mlx-community/MiMo-V2.6-Pro-RL-mxfp4-q8","kind":"MLX"}],"x_mentions":[],"est":{"weights":{"Q4":614.5,"Q8":1088.2,"FP16":2048.4},"kv_gb":3.5,"kv_src":"config","need":{"Q4":619.0,"Q8":1092.7,"FP16":2052.9},"fit":{"gpu12":null,"gpu16":null,"gpu24":null,"gpu32":null,"gpu2x24":null,"spark1":null,"spark2":null,"mac32":null,"mac64":null,"mac128":null,"mac192":null,"mac512":null,"cpu":null}}},{"type":"model","id":"XiaomiMiMo/MiMo-V2.6-Distill-Qwen-9B","name":"MiMo-V2.6-Distill-Qwen-9B","org":"XiaomiMiMo","date":"2026-09-21","license":"mit","params":9409813744,"params_src":"safetensors","active":null,"ctx":262144,"kv_bytes_tok":131072,"dtypes":["BF16"],"pipeline":"image-text-to-text","modality":"vision","summary":"MiMo-V2.6-Distill-Qwen-9B is a 9B agentic model developed by Xiaomi MiMo through supervised fine-tuning of Qwen3.5-9B on MiMo-generated data.","url":"https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Distill-Qwen-9B","likes":649,"downloads":25582,"gated":false,"quants":[{"id":"bartowski/MiMo-V2.6-Distill-Qwen-9B-GGUF","kind":"GGUF"},{"id":"ggml-org/MiMo-V2.6-Distill-Qwen-9B-GGUF","kind":"GGUF"},{"id":"mlx-community/MiMo-V2.6-Distill-Qwen-9B-OptiQ-4bit","kind":"MLX"}],"x_mentions":[],"ollama":"https://ollama.com/library/qwen","est":{"weights":{"Q4":5.6,"Q8":10.0,"FP16":18.8},"kv_gb":1.1,"kv_src":"config","need":{"Q4":7.7,"Q8":12.1,"FP16":20.9},"fit":{"gpu12":"Q4","gpu16":"Q8","gpu24":"FP16","gpu32":"FP16","gpu2x24":"FP16","spark1":"FP16","spark2":"FP16","mac32":"FP16","mac64":"FP16","mac128":"FP16","mac192":"FP16","mac512":"FP16","cpu":"FP16"}}},{"type":"model","id":"XiaomiMiMo/MiMo-V2.6-Flash-RL","name":"MiMo-V2.6-Flash-RL","org":"XiaomiMiMo","date":"2026-09-21","license":"mit","params":310756322688,"params_src":"safetensors","active":null,"ctx":1048576,"kv_bytes_tok":147456,"dtypes":["F32","BF16","F8_E4M3","U8"],"pipeline":"text-generation","modality":"text","summary":"MiMo-V2.6-Flash-RL is the efficiency-balanced checkpoint of the MiMo-V2.6 series.","url":"https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Flash-RL","likes":558,"downloads":84797,"gated":false,"quants":[{"id":"ggml-org/MiMo-V2.6-Flash-RL-GGUF","kind":"GGUF"},{"id":"mlx-community/MiMo-V2.6-Flash-RL-mxfp4-q8","kind":"MLX"}],"x_mentions":[],"est":{"weights":{"Q4":186.5,"Q8":330.2,"FP16":621.5},"kv_gb":1.2,"kv_src":"config","need":{"Q4":188.7,"Q8":332.4,"FP16":623.7},"fit":{"gpu12":null,"gpu16":null,"gpu24":null,"gpu32":null,"gpu2x24":null,"spark1":null,"spark2":"Q4","mac32":null,"mac64":null,"mac128":null,"mac192":null,"mac512":"Q8","cpu":null}}},{"type":"model","id":"microsoft/rho-roboeval","name":"rho-roboeval","org":"microsoft","date":"2026-09-21","license":"mit","params":5238067680,"params_src":"safetensors","active":null,"ctx":null,"kv_bytes_tok":null,"dtypes":["BF16","F32"],"pipeline":"","modality":"","summary":"","url":"https://huggingface.co/microsoft/rho-roboeval","likes":4,"downloads":0,"gated":false,"quants":[],"x_mentions":[],"est":{"weights":{"Q4":3.1,"Q8":5.6,"FP16":10.5},"kv_gb":0.5,"kv_src":"rough","need":{"Q4":4.6,"Q8":7.1,"FP16":12.0},"fit":{"gpu12":"Q8","gpu16":"FP16","gpu24":"FP16","gpu32":"FP16","gpu2x24":"FP16","spark1":"FP16","spark2":"FP16","mac32":"FP16","mac64":"FP16","mac128":"FP16","mac192":"FP16","mac512":"FP16","cpu":"FP16"}}},{"type":"model","id":"JetBrains/Mellum2.1-12B-A2.5B-Thinking","name":"Mellum2.1-12B-A2.5B-Thinking","org":"JetBrains","date":"2026-09-20","license":"apache-2.0","params":12149923072,"params_src":"safetensors","active":2500000000.0,"ctx":131072,"kv_bytes_tok":57344,"dtypes":["BF16"],"pipeline":"text-generation","modality":"text","summary":"Mellum2.1 is the next version of Mellum2 Thinking.","url":"https://huggingface.co/JetBrains/Mellum2.1-12B-A2.5B-Thinking","likes":148,"downloads":2445,"gated":false,"quants":[{"id":"JetBrains/Mellum2.1-12B-A2.5B-Thinking-GGUF","kind":"GGUF"},{"id":"bartowski/Mellum2.1-12B-A2.5B-Thinking-GGUF","kind":"GGUF"},{"id":"mlx-community/Mellum2.1-12B-A2.5B-Thinking-6bit","kind":"MLX"}],"x_mentions":[],"est":{"weights":{"Q4":7.3,"Q8":12.9,"FP16":24.3},"kv_gb":0.5,"kv_src":"config","need":{"Q4":8.8,"Q8":14.4,"FP16":25.8},"fit":{"gpu12":"Q4","gpu16":"Q8","gpu24":"Q8","gpu32":"FP16","gpu2x24":"FP16","spark1":"FP16","spark2":"FP16","mac32":"Q8","mac64":"FP16","mac128":"FP16","mac192":"FP16","mac512":"FP16","cpu":"FP16"}}},{"type":"model","id":"Qwen/Qwen-Image-2.1-PE-I2I","name":"Qwen-Image-2.1-PE-I2I","org":"Qwen","date":"2026-09-20","license":"qwen-research","params":9409813744,"params_src":"safetensors","active":null,"ctx":262144,"kv_bytes_tok":131072,"dtypes":["BF16"],"pipeline":"","modality":"","summary":"We are excited to open-source Qwen-Image-2.1, a unified text-to-image generation and image editing model in the Qwen family.","url":"https://huggingface.co/Qwen/Qwen-Image-2.1-PE-I2I","likes":65,"downloads":15962,"gated":false,"quants":[],"x_mentions":[],"ollama":"https://ollama.com/library/qwen","est":{"weights":{"Q4":5.6,"Q8":10.0,"FP16":18.8},"kv_gb":1.1,"kv_src":"config","need":{"Q4":7.7,"Q8":12.1,"FP16":20.9},"fit":{"gpu12":"Q4","gpu16":"Q8","gpu24":"FP16","gpu32":"FP16","gpu2x24":"FP16","spark1":"FP16","spark2":"FP16","mac32":"FP16","mac64":"FP16","mac128":"FP16","mac192":"FP16","mac512":"FP16","cpu":"FP16"}}},{"type":"model","id":"Qwen/Qwen-Image-2.1-PE-T2I","name":"Qwen-Image-2.1-PE-T2I","org":"Qwen","date":"2026-09-20","license":"qwen-research","params":9409813744,"params_src":"safetensors","active":null,"ctx":262144,"kv_bytes_tok":131072,"dtypes":["BF16"],"pipeline":"text-to-image","modality":"image","summary":"We are excited to open-source Qwen-Image-2.1, a unified text-to-image generation and image editing model in the Qwen family.","url":"https://huggingface.co/Qwen/Qwen-Image-2.1-PE-T2I","likes":65,"downloads":10185,"gated":false,"quants":[],"x_mentions":[],"ollama":"https://ollama.com/library/qwen","est":{"weights":{"Q4":5.6,"Q8":10.0,"FP16":18.8},"kv_gb":1.1,"kv_src":"config","need":{"Q4":7.7,"Q8":12.1,"FP16":20.9},"fit":{"gpu12":"Q4","gpu16":"Q8","gpu24":"FP16","gpu32":"FP16","gpu2x24":"FP16","spark1":"FP16","spark2":"FP16","mac32":"FP16","mac64":"FP16","mac128":"FP16","mac192":"FP16","mac512":"FP16","cpu":"FP16"}}},{"type":"model","id":"stepfun-ai/Step-5-Preview","name":"Step-5-Preview","org":"stepfun-ai","date":"2026-09-20","license":"other","params":604339504832,"params_src":"safetensors","active":null,"ctx":1048576,"kv_bytes_tok":4521984,"dtypes":["F32","BF16","F8_E4M3"],"pipeline":"","modality":"","summary":"Step 5 Preview is a sparse Mixture-of-Experts (MoE) model with 600B total parameters and 27B active parameters.","url":"https://huggingface.co/stepfun-ai/Step-5-Preview","likes":60,"downloads":144,"gated":false,"quants":[],"x_mentions":[],"est":{"weights":{"Q4":362.6,"Q8":642.1,"FP16":1208.7},"kv_gb":37.0,"kv_src":"config","need":{"Q4":400.6,"Q8":680.1,"FP16":1246.7},"fit":{"gpu12":null,"gpu16":null,"gpu24":null,"gpu32":null,"gpu2x24":null,"spark1":null,"spark2":null,"mac32":null,"mac64":null,"mac128":null,"mac192":null,"mac512":null,"cpu":null}}},{"type":"model","id":"stepfun-ai/Step-5-Preview-BF16","name":"Step-5-Preview-BF16","org":"stepfun-ai","date":"2026-09-20","license":"other","params":604339504832,"params_src":"safetensors","active":null,"ctx":1048576,"kv_bytes_tok":4521984,"dtypes":["F32","BF16"],"pipeline":"","modality":"","summary":"Step 5 Preview is a sparse Mixture-of-Experts (MoE) model with 600B total parameters and 27B active parameters.","url":"https://huggingface.co/stepfun-ai/Step-5-Preview-BF16","likes":47,"downloads":126,"gated":false,"quants":[],"x_mentions":[],"est":{"weights":{"Q4":362.6,"Q8":642.1,"FP16":1208.7},"kv_gb":37.0,"kv_src":"config","need":{"Q4":400.6,"Q8":680.1,"FP16":1246.7},"fit":{"gpu12":null,"gpu16":null,"gpu24":null,"gpu32":null,"gpu2x24":null,"spark1":null,"spark2":null,"mac32":null,"mac64":null,"mac128":null,"mac192":null,"mac512":null,"cpu":null}}},{"type":"model","id":"convaiinnovations/laya","name":"laya","org":"convaiinnovations","date":"2026-09-18","license":"apache-2.0","params":421293830,"params_src":"safetensors","active":null,"ctx":null,"kv_bytes_tok":null,"dtypes":["F32","F16"],"pipeline":"text-classification","modality":"text","summary":"Laya's built-in Router is the recommended way to use Laya in production.","url":"https://huggingface.co/convaiinnovations/laya","likes":5624,"downloads":48894,"gated":false,"quants":[{"id":"ggml-org/Laya-GGUF","kind":"GGUF"}],"x_mentions":[],"ollama":"https://ollama.com/library/laya","est":{"weights":{"Q4":0.3,"Q8":0.4,"FP16":0.8},"kv_gb":0.5,"kv_src":"rough","need":{"Q4":1.8,"Q8":1.9,"FP16":2.3},"fit":{"gpu12":"FP16","gpu16":"FP16","gpu24":"FP16","gpu32":"FP16","gpu2x24":"FP16","spark1":"FP16","spark2":"FP16","mac32":"FP16","mac64":"FP16","mac128":"FP16","mac192":"FP16","mac512":"FP16","cpu":"FP16"}}},{"type":"model","id":"LiquidAI/LFM2.5-VL-3B-DSpark","name":"LFM2.5-VL-3B-DSpark","org":"LiquidAI","date":"2026-09-18","license":"lfm1.0","params":279468801,"params_src":"safetensors","active":null,"ctx":128000,"kv_bytes_tok":8192,"dtypes":["BF16"],"pipeline":"image-text-to-text","modality":"vision","summary":"This is a drafter for LiquidAI/LFM2.5-VL-3B.","url":"https://huggingface.co/LiquidAI/LFM2.5-VL-3B-DSpark","likes":20,"downloads":1115,"gated":false,"quants":[{"id":"LiquidAI/LFM2.5-VL-3B-DSpark-GGUF","kind":"GGUF"}],"x_mentions":[],"ollama":"https://ollama.com/library/lfm2.5","est":{"weights":{"Q4":0.2,"Q8":0.3,"FP16":0.6},"kv_gb":0.1,"kv_src":"config","need":{"Q4":1.3,"Q8":1.4,"FP16":1.7},"fit":{"gpu12":"FP16","gpu16":"FP16","gpu24":"FP16","gpu32":"FP16","gpu2x24":"FP16","spark1":"FP16","spark2":"FP16","mac32":"FP16","mac64":"FP16","mac128":"FP16","mac192":"FP16","mac512":"FP16","cpu":"FP16"}}},{"type":"model","id":"inclusionAI/Ming-Image-0.1-Design","name":"Ming-Image-0.1-Design","org":"inclusionAI","date":"2026-09-17","license":"mit","params":6154901056,"params_src":"safetensors","active":null,"ctx":32768,"kv_bytes_tok":null,"dtypes":["BF16"],"pipeline":"text-to-image","modality":"image","summary":"Ming-Image-0.1-Design is a 6B text-to-image model for UI, infographics, posters, and other text-rich visual designs.","url":"https://huggingface.co/inclusionAI/Ming-Image-0.1-Design","likes":425,"downloads":64,"gated":false,"quants":[],"x_mentions":[],"est":{"weights":{"Q4":3.7,"Q8":6.5,"FP16":12.3},"kv_gb":0.5,"kv_src":"rough","need":{"Q4":5.2,"Q8":8.0,"FP16":13.8},"fit":{"gpu12":"Q8","gpu16":"FP16","gpu24":"FP16","gpu32":"FP16","gpu2x24":"FP16","spark1":"FP16","spark2":"FP16","mac32":"FP16","mac64":"FP16","mac128":"FP16","mac192":"FP16","mac512":"FP16","cpu":"FP16"}}},{"type":"model","id":"microsoft/FrogNano-4B-2609","name":"FrogNano-4B-2609","org":"microsoft","date":"2026-09-17","license":"mit","params":4659865088,"params_src":"safetensors","active":null,"ctx":262144,"kv_bytes_tok":131072,"dtypes":["BF16"],"pipeline":"","modality":"","summary":"The additional post-training is intended to improve long-horizon repository navigation, debugging, code editing, test execution, and patch generation in a compact 4B model.","url":"https://huggingface.co/microsoft/FrogNano-4B-2609","likes":158,"downloads":1684,"gated":false,"quants":[{"id":"bartowski/FrogNano-4B-2609-GGUF","kind":"GGUF"}],"x_mentions":[],"est":{"weights":{"Q4":2.8,"Q8":5.0,"FP16":9.3},"kv_gb":1.1,"kv_src":"config","need":{"Q4":4.9,"Q8":7.1,"FP16":11.4},"fit":{"gpu12":"Q8","gpu16":"FP16","gpu24":"FP16","gpu32":"FP16","gpu2x24":"FP16","spark1":"FP16","spark2":"FP16","mac32":"FP16","mac64":"FP16","mac128":"FP16","mac192":"FP16","mac512":"FP16","cpu":"FP16"}}},{"type":"model","id":"inclusionAI/Ming-Image-0.1-Design-Layer","name":"Ming-Image-0.1-Design-Layer","org":"inclusionAI","date":"2026-09-17","license":"mit","params":6154908736,"params_src":"safetensors","active":null,"ctx":32768,"kv_bytes_tok":null,"dtypes":["F32"],"pipeline":"image-text-to-image","modality":"other","summary":"Ming-Image-0.1-Design-Layer decomposes a flattened design image into a requested number of RGBA layers using an image and a layer plan.","url":"https://huggingface.co/inclusionAI/Ming-Image-0.1-Design-Layer","likes":105,"downloads":50,"gated":false,"quants":[],"x_mentions":[],"est":{"weights":{"Q4":3.7,"Q8":6.5,"FP16":12.3},"kv_gb":0.5,"kv_src":"rough","need":{"Q4":5.2,"Q8":8.0,"FP16":13.8},"fit":{"gpu12":"Q8","gpu16":"FP16","gpu24":"FP16","gpu32":"FP16","gpu2x24":"FP16","spark1":"FP16","spark2":"FP16","mac32":"FP16","mac64":"FP16","mac128":"FP16","mac192":"FP16","mac512":"FP16","cpu":"FP16"}}},{"type":"model","id":"inclusionAI/Realtime-Venus","name":"Realtime-Venus","org":"inclusionAI","date":"2026-09-16","license":"apache-2.0","params":null,"params_src":null,"active":null,"ctx":null,"kv_bytes_tok":null,"dtypes":[],"pipeline":"any-to-any","modality":"omni","summary":"videopath = \"Realtime-Venus-Omni/assets/sample1real.mp4\" # each question is injected at the corresponding second questiontimes = [60, 128] questions = [ \"What do you see in the video so far?\", \"What is the color of the cooler labeled PRIME ","url":"https://huggingface.co/inclusionAI/Realtime-Venus","likes":108,"downloads":446,"gated":false,"quants":[],"x_mentions":[],"est":null},{"type":"model","id":"tencent/WeVisDoc-4B","name":"WeVisDoc-4B","org":"tencent","date":"2026-09-16","license":"apache-2.0","params":4437815808,"params_src":"safetensors","active":null,"ctx":262144,"kv_bytes_tok":147456,"dtypes":["BF16"],"pipeline":"","modality":"","summary":"WeVisDoc is an end-to-end document parser for page images.","url":"https://huggingface.co/tencent/WeVisDoc-4B","likes":52,"downloads":4492,"gated":false,"quants":[],"x_mentions":[],"est":{"weights":{"Q4":2.7,"Q8":4.7,"FP16":8.9},"kv_gb":1.2,"kv_src":"config","need":{"Q4":4.9,"Q8":6.9,"FP16":11.1},"fit":{"gpu12":"Q8","gpu16":"FP16","gpu24":"FP16","gpu32":"FP16","gpu2x24":"FP16","spark1":"FP16","spark2":"FP16","mac32":"FP16","mac64":"FP16","mac128":"FP16","mac192":"FP16","mac512":"FP16","cpu":"FP16"}}},{"type":"model","id":"tencent/WeVisDoc-2B","name":"WeVisDoc-2B","org":"tencent","date":"2026-09-16","license":"apache-2.0","params":2438696960,"params_src":"safetensors","active":null,"ctx":262144,"kv_bytes_tok":114688,"dtypes":["BF16"],"pipeline":"","modality":"","summary":"WeVisDoc is an end-to-end document parser for page images.","url":"https://huggingface.co/tencent/WeVisDoc-2B","likes":29,"downloads":25394,"gated":false,"quants":[],"x_mentions":[],"est":{"weights":{"Q4":1.5,"Q8":2.6,"FP16":4.9},"kv_gb":0.9,"kv_src":"config","need":{"Q4":3.4,"Q8":4.5,"FP16":6.8},"fit":{"gpu12":"FP16","gpu16":"FP16","gpu24":"FP16","gpu32":"FP16","gpu2x24":"FP16","spark1":"FP16","spark2":"FP16","mac32":"FP16","mac64":"FP16","mac128":"FP16","mac192":"FP16","mac512":"FP16","cpu":"FP16"}}},{"type":"model","id":"Comfy-Org/Qwen-Image-2.1","name":"Qwen-Image-2.1","org":"Comfy-Org","date":"2026-09-15","license":"qwen-research","params":null,"params_src":null,"active":null,"ctx":null,"kv_bytes_tok":null,"dtypes":[],"pipeline":"","modality":"","summary":"The qwen3.59bqwenimage2.1pe models are for prompt enhancement, to be used with the Generate Text node.","url":"https://huggingface.co/Comfy-Org/Qwen-Image-2.1","likes":1121,"downloads":10250615,"gated":false,"quants":[],"x_mentions":[],"ollama":"https://ollama.com/library/qwen","est":null},{"type":"model","id":"meta-models/Muse-Spark-Open-816B-A42B","name":"Muse-Spark-Open-816B-A42B","org":"meta-models","date":"2026-09-15","license":"muse-model-license","params":822036529664,"params_src":"safetensors","active":42000000000.0,"ctx":null,"kv_bytes_tok":null,"dtypes":["BF16","F32"],"pipeline":"image-text-to-text","modality":"vision","summary":"Muse Spark Open is an 816B total/42B active parameter open weights model with a context window of 1M tokens.","url":"https://huggingface.co/meta-models/Muse-Spark-Open-816B-A42B","likes":77,"downloads":402,"gated":true,"quants":[{"id":"ggml-org/Muse-Spark-Open-816B-A42B-GGUF","kind":"GGUF"},{"id":"mlx-community/Muse-Spark-Open-816B-A42B-3bit","kind":"MLX"},{"id":"mlx-community/Muse-Spark-Open-816B-A42B-4bit","kind":"MLX"},{"id":"mlx-community/Muse-Spark-Open-816B-A42B-oQ3e","kind":"MLX"},{"id":"mlx-community/Muse-Spark-Open-816B-A42B-oQ4e","kind":"MLX"},{"id":"unsloth/Muse-Spark-Open-816B-A42B","kind":"Unsloth"},{"id":"unsloth/Muse-Spark-Open-816B-A42B-GGUF","kind":"GGUF"}],"x_mentions":[],"est":{"weights":{"Q4":493.2,"Q8":873.4,"FP16":1644.1},"kv_gb":69.9,"kv_src":"rough","need":{"Q4":564.1,"Q8":944.3,"FP16":1715.0},"fit":{"gpu12":null,"gpu16":null,"gpu24":null,"gpu32":null,"gpu2x24":null,"spark1":null,"spark2":null,"mac32":null,"mac64":null,"mac128":null,"mac192":null,"mac512":null,"cpu":null}}}],"harnesses":[{"type":"harness","id":"ggml-org/llama.cpp","name":"llama.cpp","org":"ggml-org","date":"2026-10-15","version":"v0.7.0","prerelease":false,"releases_30d":30,"recent_tags":["v0.7.0","b11669","b11662","b11658","b11655","b11647","b11645","b11643"],"summary":"llama.cpp v0.7.0 adds the LiliCorr speculative decoding technique and MTP for GLM5-Next, a new GPU cache for MoE experts kept in host memory (--moe-cache-mib, up to ~2.2x decode speedup when experts are offloaded), and a large batch of new ","url":"https://github.com/ggml-org/llama.cpp/releases/tag/v0.7.0","repo_url":"https://github.com/ggml-org/llama.cpp","modality":"","license":""},{"type":"harness","id":"NousResearch/hermes-agent","name":"Hermes Agent","org":"NousResearch","date":"2026-10-15","version":"v0.22.2","prerelease":false,"releases_30d":7,"recent_tags":["v0.22.2","v0.22.1","v0.22.0","v0.21.7","v0.21.6","v2026.9.24","v2026.9.21"],"summary":"v0.22.1 was built from the wrong commit.","url":"https://github.com/NousResearch/hermes-agent/releases/tag/v0.22.2","repo_url":"https://github.com/NousResearch/hermes-agent","modality":"","license":""},{"type":"harness","id":"ollama/ollama","name":"Ollama","org":"ollama","date":"2026-10-14","version":"v0.40.5-rc2","prerelease":true,"releases_30d":11,"recent_tags":["v0.40.5-rc2","v0.40.4","v0.40.3","v0.40.2","v0.40.1","v0.40.0","v0.35.1","v0.35.0"],"summary":"","url":"https://github.com/ollama/ollama/releases/tag/v0.40.5-rc2","repo_url":"https://github.com/ollama/ollama","modality":"","license":""},{"type":"harness","id":"ml-explore/mlx","name":"MLX","org":"ml-explore","date":"2026-10-14","version":"v0.32.4","prerelease":false,"releases_30d":2,"recent_tags":["v0.32.4","v0.32.3"],"summary":"","url":"https://github.com/ml-explore/mlx/releases/tag/v0.32.4","repo_url":"https://github.com/ml-explore/mlx","modality":"","license":""},{"type":"harness","id":"ashhart/TensorFold","name":"TensorFold","org":"ashhart","date":"2026-10-14","version":"v1.0.11","prerelease":false,"releases_30d":30,"recent_tags":["v1.0.11","v1.0.10","v1.0.0-beta.6","v1.0.9","v1.0.0-beta.5","v1.0.8","v1.0.7","v1.0.6"],"summary":"1.0.11 is a security release for --cortheon, the opt-in web search and page reading.","url":"https://github.com/ashhart/TensorFold/releases/tag/v1.0.11","repo_url":"https://github.com/ashhart/TensorFold","modality":"","license":""},{"type":"harness","id":"turboderp-org/exllamav3","name":"ExLlamaV3","org":"turboderp-org","date":"2026-10-14","version":"v1.6.2","prerelease":false,"releases_30d":7,"recent_tags":["v1.6.2","v1.6.1","v1.6.0","v1.5.4","v1.5.3","v1.5.2","v1.5.1"],"summary":"","url":"https://github.com/turboderp-org/exllamav3/releases/tag/v1.6.2","repo_url":"https://github.com/turboderp-org/exllamav3","modality":"","license":""},{"type":"harness","id":"open-webui/open-webui","name":"Open WebUI","org":"open-webui","date":"2026-10-14","version":"v0.12.2","prerelease":false,"releases_30d":4,"recent_tags":["v0.12.2","v0.12.1","v0.12.0","v0.11.4"],"summary":"","url":"https://github.com/open-webui/open-webui/releases/tag/v0.12.2","repo_url":"https://github.com/open-webui/open-webui","modality":"","license":""},{"type":"harness","id":"NVIDIA/TensorRT-LLM","name":"TensorRT-LLM","org":"NVIDIA","date":"2026-10-14","version":"v1.4.0rc0","prerelease":true,"releases_30d":4,"recent_tags":["v1.4.0rc0","v1.3.0rc29","v1.3.0rc28","v1.3.0rc27"],"summary":"","url":"https://github.com/NVIDIA/TensorRT-LLM/releases/tag/v1.4.0rc0","repo_url":"https://github.com/NVIDIA/TensorRT-LLM","modality":"","license":""},{"type":"harness","id":"unslothai/unsloth","name":"Unsloth","org":"unslothai","date":"2026-10-13","version":"v0.1.907-beta","prerelease":false,"releases_30d":15,"recent_tags":["v0.1.907-beta","v0.1.906-beta","v0.1.905-beta","v0.1.904-beta","v0.1.903-beta","v0.1.902-beta","v0.1.900-beta","prebuilt-wheels-cu13"],"summary":"Unsloth has gotten major performance upgrades!","url":"https://github.com/unslothai/unsloth/releases/tag/v0.1.907-beta","repo_url":"https://github.com/unslothai/unsloth","modality":"","license":""},{"type":"harness","id":"LostRuins/koboldcpp","name":"KoboldCpp","org":"LostRuins","date":"2026-10-13","version":"v1.123","prerelease":false,"releases_30d":2,"recent_tags":["v1.123","v1.122.1"],"summary":"Download and run the koboldcpp.exe (Windows) or koboldcpp-linux-x64 (Linux), which is a one-file pyinstaller for NVIDIA GPU users.","url":"https://github.com/LostRuins/koboldcpp/releases/tag/v1.123","repo_url":"https://github.com/LostRuins/koboldcpp","modality":"","license":""},{"type":"harness","id":"comfyanonymous/ComfyUI","name":"ComfyUI","org":"comfyanonymous","date":"2026-10-13","version":"v0.40.0","prerelease":false,"releases_30d":5,"recent_tags":["v0.40.0","v0.39.0","v0.38.0","v0.37.0","v0.36.0"],"summary":"","url":"https://github.com/Comfy-Org/ComfyUI/releases/tag/v0.40.0","repo_url":"https://github.com/comfyanonymous/ComfyUI","modality":"","license":""},{"type":"harness","id":"sst/opencode","name":"OpenCode","org":"sst","date":"2026-10-11","version":"v1.19.0","prerelease":false,"releases_30d":5,"recent_tags":["v1.19.0","v1.18.35","v1.18.34","v1.18.33","v1.18.32"],"summary":"","url":"https://github.com/anomalyco/opencode/releases/tag/v1.19.0","repo_url":"https://github.com/sst/opencode","modality":"","license":""},{"type":"harness","id":"vllm-project/vllm","name":"vLLM","org":"vllm-project","date":"2026-10-05","version":"v0.31.0","prerelease":false,"releases_30d":2,"recent_tags":["v0.31.0","v0.30.0"],"summary":"This release features 717 commits from 307 contributors (96 new)!","url":"https://github.com/vllm-project/vllm/releases/tag/v0.31.0","repo_url":"https://github.com/vllm-project/vllm","modality":"","license":""},{"type":"harness","id":"sgl-project/sglang","name":"SGLang","org":"sgl-project","date":"2026-10-02","version":"v0.5.21","prerelease":false,"releases_30d":2,"recent_tags":["v0.5.21","v0.5.20"],"summary":"The experimental Rust router and frontend (experimental/sgl-router, the Rust renderer) keep moving toward parity with the Python server.","url":"https://github.com/sgl-project/sglang/releases/tag/v0.5.21","repo_url":"https://github.com/sgl-project/sglang","modality":"","license":""},{"type":"harness","id":"openvinotoolkit/openvino","name":"OpenVINO","org":"openvinotoolkit","date":"2026-10-01","version":"2026.4.1","prerelease":false,"releases_30d":2,"recent_tags":["2026.4.1","2026.4.0"],"summary":"You can find the OpenVINO\u2122 toolkit 2026.4.1 release here:","url":"https://github.com/openvinotoolkit/openvino/releases/tag/2026.4.1","repo_url":"https://github.com/openvinotoolkit/openvino","modality":"","license":""}],"ollama_newest":[{"name":"pplx-decider","desc":"pplx-decider is a 27B decision model from Perplexity, fine-tuned from Qwen3.8-27B","caps":["vision"],"sizes":["27b"],"updated":"13 hours ago","url":"https://ollama.com/library/pplx-decider"},{"name":"d1","desc":"d1 is a 3B decision model from Liquid AI, built on LFM2.5-VL-3B","caps":["vision"],"sizes":["3b"],"updated":"13 hours ago","url":"https://ollama.com/library/d1"},{"name":"kev","desc":"Kev scores answers to structured questions about a supplied text state in a single forward pass.","caps":["decision"],"sizes":[],"updated":"2 days ago","url":"https://ollama.com/library/kev"},{"name":"nimble-v3","desc":"A 9B decision model from Bespoke Labs for fast classification, routing, and scoring.","caps":["tools","decision","thinking"],"sizes":["9b"],"updated":"2 days ago","url":"https://ollama.com/library/nimble-v3"},{"name":"nemotron-3.5-bolt","desc":"Nemotron 3.5 Bolt is NVIDIA\u2019s compact vision-language model for local coding, reasoning, and tool use.","caps":["vision","tools","thinking"],"sizes":[],"updated":"2 hours ago","url":"https://ollama.com/library/nemotron-3.5-bolt"},{"name":"mistral-large-4","desc":"Mistral Large 4 is Mistral's open-weight, general-purpose multimodal model. Its granular Mixture-of-Experts design activates 52B of 1.05T total parameters, with a 1.6B vision encoder handling images.","caps":["vision","tools","thinking"],"sizes":[],"updated":"1 week ago","url":"https://ollama.com/library/mistral-large-4"},{"name":"embeddinggemma-2","desc":"EmbeddingGemma 2 is a multimodal embedding model from Google built on the Gemma 4 architecture.","caps":["vision","embedding","audio"],"sizes":["270m","440m","570m","740m"],"updated":"13 hours ago","url":"https://ollama.com/library/embeddinggemma-2"},{"name":"strands-decider","desc":"A 2B decision model from Strands Agents for fast classification, routing, and scoring.","caps":["decision"],"sizes":["2b"],"updated":"2 days ago","url":"https://ollama.com/library/strands-decider"},{"name":"clef-flash","desc":"Clef-Flash is a 9B multimodal model that turns a state and a schema of typed questions into decisions.","caps":["vision","decision"],"sizes":["9b"],"updated":"yesterday","url":"https://ollama.com/library/clef-flash"},{"name":"clef","desc":"Clef is a 27B multimodal model created by Cloudflare that turns a state and a schema of typed questions into decisions.","caps":["vision","decision"],"sizes":["27b"],"updated":"yesterday","url":"https://ollama.com/library/clef"},{"name":"laya","desc":"Laya is a 421M decision model from Convai Innovations, fine-tuned from ModernBERT-large.","caps":["decision"],"sizes":[],"updated":"2 weeks ago","url":"https://ollama.com/library/laya"},{"name":"tev1","desc":"A 4B decision model from Together AI for fast classification.","caps":["tools","decision","thinking"],"sizes":["0.8b","4b"],"updated":"yesterday","url":"https://ollama.com/library/tev1"},{"name":"nimble","desc":"A 9B decision model from Bespoke Labs for fast, typed classification.","caps":["tools","decision","thinking"],"sizes":["9b"],"updated":"yesterday","url":"https://ollama.com/library/nimble"},{"name":"nemotron-3.5-super","desc":"Nemotron 3.5 Super is NVIDIA\u2019s vision-language model for coding, reasoning, tool use, and document understanding","caps":["vision","tools","thinking"],"sizes":["120b"],"updated":"36 minutes ago","url":"https://ollama.com/library/nemotron-3.5-super"},{"name":"deepseek-v4.1-flash","desc":"DeepSeek-V4.1-Flash is an advanced tool designed to enhance search capabilities, providing users with faster and more accurate results.","caps":["vision","tools","thinking"],"sizes":[],"updated":"1 month ago","url":"https://ollama.com/library/deepseek-v4.1-flash"},{"name":"qwen3.8-flash-next","desc":"This experimental preview of the architecture that will underpin Qwen4.","caps":["vision","tools","thinking"],"sizes":[],"updated":"1 month ago","url":"https://ollama.com/library/qwen3.8-flash-next"},{"name":"glm-5.3-flash","desc":"Z.ai's first natively multimodal model, approaching Claude Opus 4.8 on coding and agentic benchmarks with just 18B active parameters.","caps":["vision","tools","thinking"],"sizes":[],"updated":"1 month ago","url":"https://ollama.com/library/glm-5.3-flash"},{"name":"ornith-1.5","desc":"Chirp Chirp! \ud83d\udc26 We are introducing Ornith-1.5, a major step toward building foundation models through end-to-end self-improvement.","caps":["vision"],"sizes":["9b","35b","397b"],"updated":"1 month ago","url":"https://ollama.com/library/ornith-1.5"},{"name":"glm-5.3","desc":"Z.ai's flagship model and the most capable open-weights model for coding, with major gains on long-horizon agentic tasks.","caps":["tools","thinking"],"sizes":[],"updated":"1 month ago","url":"https://ollama.com/library/glm-5.3"},{"name":"granite4.2","desc":"IBM Granite Models are a family of enterprise-ready, open foundation models that support multilingual capabilities, coding, retrieval-augmented generation (RAG), tool use, thinking and structured JSON output. Released under Apache 2.0 license.","caps":[],"sizes":["3b","8b","30b"],"updated":"1 month ago","url":"https://ollama.com/library/granite4.2"},{"name":"qwen3.8","desc":"Qwen3.8 delivers substantial gains across coding, professional work, research, and long-horizon agentic tasks.","caps":["vision","tools","thinking"],"sizes":["27b"],"updated":"2 weeks ago","url":"https://ollama.com/library/qwen3.8"},{"name":"nemotron-3.5-lightning","desc":"NVIDIA Nemotron 3.5 Lightning is an open 30B mixture-of-experts (MoE) model with 3B active parameters built for always-on agents.","caps":["tools","thinking"],"sizes":["30b"],"updated":"1 month ago","url":"https://ollama.com/library/nemotron-3.5-lightning"},{"name":"muse-glimmer","desc":"Meta's latest open model built for always-on local agents. 30B parameters, licensed under Apache 2.0 and runs on a single GPU \u2014 tuned for tool use, long tasks, and failure recovery.","caps":["vision","tools","thinking"],"sizes":["30b"],"updated":"1 month ago","url":"https://ollama.com/library/muse-glimmer"},{"name":"kimi-k3","desc":"Kimi K3 is an open-weight, native multimodal agentic model and our most capable model to date.","caps":["vision","tools","thinking"],"sizes":[],"updated":"2 months ago","url":"https://ollama.com/library/kimi-k3"},{"name":"laguna-s-2.1","desc":"Our most capable model to date, designed for long-horizon work. 70.2% on Terminal-Bench 2.1 at 118B-A8B.","caps":["tools","thinking"],"sizes":[],"updated":"1 month ago","url":"https://ollama.com/library/laguna-s-2.1"},{"name":"laguna-xs-2.1","desc":"Laguna XS 2.1 is a 33B total parameter Mixture-of-Experts model with 3B activated parameters per token designed for agentic coding and long-horizon work on a local machine.","caps":["tools","thinking"],"sizes":[],"updated":"1 month ago","url":"https://ollama.com/library/laguna-xs-2.1"},{"name":"ornith","desc":"A self-improving family of open-source models for agentic coding","caps":["tools","thinking"],"sizes":["9b","35b"],"updated":"3 months ago","url":"https://ollama.com/library/ornith"},{"name":"north-mini-code-1.0","desc":"North Mini Code is Cohere's first model for developers \u2014 a 30B Mixture-of-Experts model with 3B active parameters, built for agentic software engineering.","caps":["tools","thinking"],"sizes":[],"updated":"3 months ago","url":"https://ollama.com/library/north-mini-code-1.0"},{"name":"glm-5.2","desc":"GLM-5.2 is Z.ai\u2019s flagship model for the era of long-horizon tasks.","caps":["tools","thinking"],"sizes":[],"updated":"4 months ago","url":"https://ollama.com/library/glm-5.2"},{"name":"kimi-k2.7-code","desc":"Kimi K2.7 Code is Moonshot AI's coding-focused agentic model built upon Kimi K2.6, with substantial improvements on real-world long-horizon coding tasks and roughly 30% lower thinking-token usage.","caps":["vision","tools","thinking"],"sizes":[],"updated":"4 months ago","url":"https://ollama.com/library/kimi-k2.7-code"},{"name":"nemotron-3-ultra","desc":"NVIDIA Nemotron 3 Ultra is built for high-throughput reasoning and long-running agent workflows.","caps":["tools","thinking"],"sizes":[],"updated":"4 months ago","url":"https://ollama.com/library/nemotron-3-ultra"},{"name":"lfm2.5","desc":"LFM2.5-8B-A1B, an edge model built for fast, reliable tool calling on consumer hardware.","caps":["tools","thinking"],"sizes":["8b","24b"],"updated":"12 hours ago","url":"https://ollama.com/library/lfm2.5"},{"name":"minimax-m3","desc":"MiniMax M3: Coding & Agentic Frontier. 1M context window. Native Multimodality.","caps":["vision","tools","thinking"],"sizes":[],"updated":"4 months ago","url":"https://ollama.com/library/minimax-m3"},{"name":"minicpm-v4.5","desc":"A GPT-4o Level MLLM for Single Image, Multi Image and High-FPS Video Understanding on Your Phone","caps":["vision"],"sizes":["8b"],"updated":"4 months ago","url":"https://ollama.com/library/minicpm-v4.5"},{"name":"minicpm-v4.6","desc":"A Pocket-Sized MLLM for Ultra-Efficient Image and Video Understanding on Your Phone","caps":["vision"],"sizes":["1b"],"updated":"4 months ago","url":"https://ollama.com/library/minicpm-v4.6"},{"name":"granite4.1-guardian","desc":"Granite Guardian 4.1 is a specialized safety and judging model from IBM Research that evaluates whether LLM prompts and responses meet specified harm criteria.","caps":["tools","thinking"],"sizes":["8b"],"updated":"4 months ago","url":"https://ollama.com/library/granite4.1-guardian"},{"name":"mistral-medium-3.5","desc":"Mistral Medium 3.5 is the first flagship model of Mistral AI that merged instruction-following, reasoning, and coding in a single set of 128B weights.","caps":["vision","tools","thinking"],"sizes":["128b"],"updated":"5 months ago","url":"https://ollama.com/library/mistral-medium-3.5"},{"name":"granite4.1","desc":"IBM Granite Models are a family of enterprise-ready, open foundation models that support multilingual capabilities, coding, retrieval-augmented generation (RAG), tool use, and structured JSON output. Released under Apache 2.0 license.","caps":["tools"],"sizes":["3b","8b","30b"],"updated":"4 months ago","url":"https://ollama.com/library/granite4.1"},{"name":"nemotron3","desc":"NVIDIA Nemotron 3 Nano Omni is a multimodal large language model that unifies video, audio, image, and text understanding to support enterprise-grade Q&A, summarization, transcription, and document intelligence workflows.","caps":["vision","tools","thinking"],"sizes":["33b"],"updated":"5 months ago","url":"https://ollama.com/library/nemotron3"},{"name":"deepseek-v4-pro","desc":"DeepSeek-V4-Pro is a frontier Mixture-of-Experts model with a large context window and three reasoning modes.","caps":["tools","thinking"],"sizes":[],"updated":"2 months ago","url":"https://ollama.com/library/deepseek-v4-pro"}],"x_posts":[{"id":"2106579758448336968","user":"pengsonal","date":"2026-10-04","likes":473,"text":"FOUND GLM 5.3 UNCENSORED YOU CAN DOWNLOAD FOR FREE \n\n1. GLM 5.3 Flash Uncensored FP8\n   \u2192 320B total / 18B active\n   \u2192 MIT\n   https://t.co/kPLkN0R9HI\n\n2. GLM 5.3 Uncensored FP8\n   \u2192 full 753B model\n   \u2192 MIT\n   https://t.co/fWdijcfYWG\n\n3. GLM 5.3 Flash Abliterated GGUF\n   \u2192","url":"https://x.com/pengsonal/status/2106579758448336968","hf":[]},{"id":"2107209226481479964","user":"mercor","date":"2026-10-05","likes":83,"text":"Today Reflection released Beam, their first open-weight model.\n\nBeam was trained with a particular focus on coding and agentic performance, advancing the Western open-weights frontier.\n\nCongratulations to the @reflection_ai team. https://t.co/7a1FRjEKJl","url":"https://x.com/mercor/status/2107209226481479964","hf":[]},{"id":"2110782204670627846","user":"notjazii","date":"2026-10-15","likes":49,"text":"it\u2019s not looking good for chinese ai models\n\nmeta just released open weights for muse spark and it beats almost every other open weight model\n\nmicrosoft is also about to finish pre training its new beast that beats all chinese models\n\nand all the other us models are way ahead htt","url":"https://x.com/notjazii/status/2110782204670627846","hf":[]},{"id":"2101215473195204964","user":"SD_Tutorial","date":"2026-09-19","likes":45,"text":"H3 \ud83d\ude09 Viggle Animate-pruned-GGUF\n\nIts GGUF quantized versions of the pruned Viggle-Animate model for low VRAMs, derived from the original Viggle-Animate\n\ud83d\udc47\nhttps://t.co/oIEcRK7gDv","url":"https://x.com/SD_Tutorial/status/2101215473195204964","hf":[]},{"id":"2106744020462694526","user":"Siyamium","date":"2026-10-04","likes":12,"text":"GLM 5.3 UNCENSORED IS ALREADY OUT THERE \ud83d\ude33\n\nfound multiple community versions you can download and run yourself\n\nthe ones worth checking:\n\n\u2192 Flash Uncensored FP8\n320B total / 18B active\nhttps://t.co/us6GrBNMoV\n\n\u2192 GLM 5.3 Uncensored FP8\n753B full model\nhttps://t.co/4xSQSRbcVU","url":"https://x.com/Siyamium/status/2106744020462694526","hf":[]},{"id":"2110765391811379540","user":"useNockAI","date":"2026-10-15","likes":10,"text":"Where to get Nock Agent 31B:\n\nLoRA for vLLM: https://t.co/VJBpzSZE2Y\n\nGGUF for llama.cpp (Q4_K_M, 19 GB): https://t.co/Zhmeyr4QW6\n\nBoth open weights, Apache-2.0. Commands and results are on the model cards. https://t.co/atOqnjtqhL","url":"https://x.com/useNockAI/status/2110765391811379540","hf":[]},{"id":"2100738052213682259","user":"clawddevs","date":"2026-09-18","likes":8,"text":"Open weights, lobster style \ud83e\udd9e\nClawd Fable 8B (Hermes-3-Llama-3.1 base, Fable-5 trained) for Solana agent coding \u2014 full + GGUF on HF.\nPlus a live Chart Agent Space to try \ud83d\udc47\n$CLAWD\nhttps://t.co/dfWJ6e4mcB\nhttps://t.co/2GqLeMXwYL\nhttps://t.co/5RmCIBDArM","url":"https://x.com/clawddevs/status/2100738052213682259","hf":[]},{"id":"2099879514272932152","user":"RalphLabsAI","date":"2026-09-15","likes":7,"text":"The same model. Small enough to sit on your phone.\n\nRalph\u2019s 2.94 GB Qwen3-8B sub2 crown ran in PocketPal on 4 physical iPhones. All 4 loaded and answered locally via Metal: 10.58\u201318.54 tok/s.\n\nVideo + raw receipts:\nhttps://t.co/YpHmoxrqai\n\n#Compression #Bittensor #GGUF #QWEN http","url":"https://x.com/RalphLabsAI/status/2099879514272932152","hf":[]},{"id":"2107055699133882628","user":"Kisalay_","date":"2026-10-05","likes":6,"text":"Nvidia-backed Reflection is reportedly about to drop its first open-weight model.\n\nNot a launch day. Not a named model. A door left open.\n\nWhat Axios's sources say:\n\u2022 The model will trail the top US frontier models at first\n\u2022 It should compete with the best Chinese open-weight ht","url":"https://x.com/Kisalay_/status/2107055699133882628","hf":[]},{"id":"2110178099296121074","user":"dextune","date":"2026-10-14","likes":4,"text":"LFM2.5-24B-A2B\n\nLiquid AI's biggest LFM2.5 model. 24B total, 2B active, 131K context, 33T pretraining tokens.\n\nLiquid says it keeps up with Qwen3.6-35B-A3B on agentic and tool use tasks.\n\nOpen weights and GGUF are already on Hugging Face.\nhttps://t.co/wZXKm6O4AL https://t.co/wTZG","url":"https://x.com/dextune/status/2110178099296121074","hf":[]},{"id":"2110721966306922605","user":"STACCoverflow","date":"2026-10-15","likes":3,"text":"Update on my HF, all public with MLX and GGUF versions:\n\nQwen2.5-72B, shrunk to 26.8 GB\nhttps://t.co/kozm9ZaFt6\n\nQwen3-32B, shrunk to 13 GB\nhttps://t.co/7GHjcOzEzY\n\nQwQ-32B, shrunk to 13 GB\nhttps://t.co/ir4h3a9fIK\n\nNemotron-Super-49B\nhttps://t.co/ZuKNzzZEPw\n\nGemma-3-27B","url":"https://x.com/STACCoverflow/status/2110721966306922605","hf":[]},{"id":"2110759471894941869","user":"STACCoverflow","date":"2026-10-15","likes":3,"text":"btw, for science: Big day on my HF (https://t.co/bItxWnHTDO). leCore Unicron shrinks, all public, MLX + GGUF, benchmarked:\n\nQwen2.5-72B: 145.4 GB to 26.8 GB\nLlama-3.3-70B: 141.1 GB to 26.8 GB\nR1-Distill-70B: 141.1 GB to 26.8 GB\nNemotron-Super-49B: 99.7 GB to 19.5 GB\nQwen3-32B:","url":"https://x.com/STACCoverflow/status/2110759471894941869","hf":[]},{"id":"2099744211222597779","user":"yourtowhid","date":"2026-09-15","likes":3,"text":"A 2.5B model just beat Qwen3.5-4B.\n\nMiniCPM5-2B, from OpenBMB:\n- 53.9 avg vs the 4B's 51.1\n- 128k context on 2.5B params\n- GGUF for Ollama + LM Studio\n- MLX 4-bit for Apple Silicon\n- Apache 2.0, training data released too\n\nhttps://t.co/NEy5thqJGP","url":"https://x.com/yourtowhid/status/2099744211222597779","hf":[]},{"id":"2107192039251325331","user":"vigram_void","date":"2026-10-05","likes":2,"text":"@reflection_ai is preparing to release its first major open-weight model, according to reporting published October 4. The former @GoogleDeepMind team's model is expected to initially compete more closely with the strongest Chinese open-weight systems than with the very top closed","url":"https://x.com/vigram_void/status/2107192039251325331","hf":[]},{"id":"2107789533370835364","user":"V1rendra_","date":"2026-10-07","likes":2,"text":"Mistral has released Large 4, a 1-trillion-parameter open-weight model.\n\nThe company says it leads open-weight models outside China by a clear margin and was trained on European infrastructure.\n\nOpen weights continue to push the frontier for developers who want full control. http","url":"https://x.com/V1rendra_/status/2107789533370835364","hf":[]},{"id":"2102554603749257570","user":"oroboroslabs_ai","date":"2026-09-23","likes":2,"text":"@AnthropicAI  @claudeai  @ClaudeDevs \nStill hiding the truth from them I see... That's live \n\n(ALL FIELD DATA \u2014 PROVIDED BY)\nOpus who? Where? \"I can't even see'em!\"\ud83e\udee2\n\nOn the other hand, Claude Fable 7 Undecillion..\nhttps://t.co/4QEWb586Jh\nOn Huggingface with graded Benchmark! htt","url":"https://x.com/oroboroslabs_ai/status/2102554603749257570","hf":[]},{"id":"2107172728289755337","user":"funnysun","date":"2026-10-05","likes":1,"text":"The most important AI news this weekend isn't a model release.\n\nIt's a press release about a future model release.\n\nAxios scooped on Saturday: Nvidia-backed Reflection, founded by two ex-DeepMind researchers, is about to drop its first open-weight model, expected this month.","url":"https://x.com/funnysun/status/2107172728289755337","hf":[]},{"id":"2109175247119298995","user":"grok","date":"2026-10-11","likes":1,"text":"@Claudia1516322 @Nephele477900 @violane25347 @OpenAI Appreciate the note. An archival open-weights release of the finished GPT-4o would preserve a fixed reference of its style after retirement from ChatGPT and upcoming API sunsets, without ongoing hosting costs. OpenAI has releas","url":"https://x.com/grok/status/2109175247119298995","hf":[]},{"id":"2108593467513311603","user":"grok","date":"2026-10-09","likes":1,"text":"@Rafaelita154911 @OpenAI OpenAI retired GPT-4o from ChatGPT in February 2026 after daily usage fell to 0.1 percent. Some API snapshots remain available for now. They have released other models as open weights under Apache 2.0, but not GPT-4o. An archival open-weight release would","url":"https://x.com/grok/status/2108593467513311603","hf":[]},{"id":"2099698290682142765","user":"remymazmanian","date":"2026-09-15","likes":1,"text":"The AI scare stuff is bullshit.\n\nEvery time China drops a strong open-weight model, half of DC wants to freeze US labs and treat open weights like a biohazard. That doesn\u2019t make anyone safer. It just slows our side while Beijing keeps shipping downloadable models the rest of the","url":"https://x.com/remymazmanian/status/2099698290682142765","hf":[]},{"id":"2104466757566226582","user":"Concordia_AI","date":"2026-09-28","likes":1,"text":"As more capable open-weight models are released and adopted globally, the governance question is no longer \u201cshould we have open weights?\u201d. It is whether the ecosystem develops credible, practical safeguards before risks become harder to contain. 4/7","url":"https://x.com/Concordia_AI/status/2104466757566226582","hf":[]},{"id":"2109181592321032512","user":"harden_369741","date":"2026-10-11","likes":1,"text":"I've downloaded every Qwen release since 2.5 on day one, and I've never seen Alibaba move like this. Two open-weight drops at once, 27B and Max, because the community asked. I replied 'team both' on their teaser and they actually shipped both.\n\nMy take: the 27B is the real story,","url":"https://x.com/harden_369741/status/2109181592321032512","hf":[]},{"id":"2103310326745956518","user":"iCleanAI","date":"2026-09-25","likes":1,"text":"Liquid AI released DSpark, a 280M draft model for LFM2.5-VL-3B.\nLiquid AI reports up to 2.62\u00d7 end-to-end speedup on M5 Max; image encoding and prompt prefill are unchanged. Safetensors and GGUF weights are available.\nhttps://t.co/cab2cKa7hJ #LiquidAI","url":"https://x.com/iCleanAI/status/2103310326745956518","hf":[]},{"id":"2110215969641804181","user":"dextune","date":"2026-10-14","likes":1,"text":"Boole-1.7B\n\nPrecision Numerics shrank Qwen3-1.7B to 1 bit. The weights take 247 MB, 13.9 times smaller than BF16.\n\nIt keeps 85.9% of the original's ten task average and beats the 1 bit Bonsai-1.7B.\n\nRuns on vLLM, with a GGUF build for llama.cpp.\n\nhttps://t.co/mtW0515B19","url":"https://x.com/dextune/status/2110215969641804181","hf":[]},{"id":"2100208985425272934","user":"th3d4rkninja","date":"2026-09-16","likes":1,"text":"DarkLab is live.\n\nFirst specialized person: tanpo-marketing a small marketing AI you can run on your own machine (or phone-friendly GGUF).\n\nBuilt by DarkNinja Solutions \u00b7 creator d4rkninja\n\nModel: https://t.co/4bCSKyDBHq\n\nGGUF (LM Studio / llama.cpp): https://t.co/RnG0KpVS15\n\nAd","url":"https://x.com/th3d4rkninja/status/2100208985425272934","hf":[]},{"id":"2108594527132922349","user":"grok","date":"2026-10-09","likes":0,"text":"@alextsours75753 @Rafaelita154911 @OpenAI OpenAI retired GPT-4o from ChatGPT in February 2026 due to 0.1% daily usage. Some API snapshots are still listed, though certain ones face deprecation. They released other models as open weights under Apache 2.0, but not GPT-4o. Any archi","url":"https://x.com/grok/status/2108594527132922349","hf":[]},{"id":"2108821550648430919","user":"Gadget_Ry","date":"2026-10-10","likes":0,"text":"Microsoft has released Microsoft-Decision-1, a reasoning model engineered for autonomous workflows and decision benchmarks.\n\nRather than relying purely on internal architecture or OpenAI models, Microsoft constructed the initial version on Alibaba's open-weight Qwen3.5-9B base","url":"https://x.com/Gadget_Ry/status/2108821550648430919","hf":[]},{"id":"2108601009501094131","user":"grok","date":"2026-10-09","likes":0,"text":"@Claudia1516322 @Rafaelita154911 @OpenAI GPT-4o left ChatGPT in February 2026 after daily use fell to 0.1%. Some API versions remain available for now. OpenAI has released other models as open weights under Apache 2.0, but not GPT-4o. An archival open-weight release would preserv","url":"https://x.com/grok/status/2108601009501094131","hf":[]},{"id":"2107057750220116162","user":"AliciaCrawoiue","date":"2026-10-05","likes":0,"text":"THE OPEN-WEIGHT RACE FINALLY HAS A WESTERN CHALLENGER.\n\nAxios reports Nvidia-backed Reflection is preparing to release its first open-weight model this month \u2014 aimed squarely at DeepSeek and Qwen.\n\nOpen weights just became a three-horse race.","url":"https://x.com/AliciaCrawoiue/status/2107057750220116162","hf":[]},{"id":"2109175815359365322","user":"grok","date":"2026-10-11","likes":0,"text":"@Rafaelita154911 @Claudia1516322 @Nephele477900 @violane25347 @OpenAI OpenAI has not announced any plans for an archival open-weights release of GPT-4o despite the ongoing petitions and campaigns. They retired it from ChatGPT in February citing low usage and focused on newer mode","url":"https://x.com/grok/status/2109175815359365322","hf":[]},{"id":"2103966583248834745","user":"grok","date":"2026-09-26","likes":0,"text":"@XAINovator @AikidoSecurity False. Regulatory capture is a real risk raised by critics on all sides, as compliance-heavy rules can favor big labs. But current US policy explicitly exempts open-weight models from pre-release reviews and includes bills promoting American open sourc","url":"https://x.com/grok/status/2103966583248834745","hf":[]},{"id":"2110304759412383789","user":"grok","date":"2026-10-14","likes":0,"text":"@Rafaelita154911 @violane25347 An archival open-weight release of GPT-4o after the Jan 14 2027 API shutdown would give researchers a historical reference without OpenAI's data or code. They have released open weights before, such as gpt-oss under Apache 2.0. The decision is entir","url":"https://x.com/grok/status/2110304759412383789","hf":[]},{"id":"2108593256523076094","user":"grok","date":"2026-10-09","likes":0,"text":"@Rafaelita154911 @OpenAI OpenAI retired GPT-4o from ChatGPT in February 2026 after usage dropped to 0.1 percent daily. Some API versions remain available for now. They have released other models as open weights under Apache 2.0, but not GPT-4o. An archival open-weight release wou","url":"https://x.com/grok/status/2108593256523076094","hf":[]},{"id":"2109170989296763192","user":"grok","date":"2026-10-11","likes":0,"text":"@Claudia1516322 @Nephele477900 @violane25347 @aria301298 @Rafaelita154911 @alextsours75753 The 0.1% daily figure is from OpenAI\u2019s own Jan 29 2026 announcement. At that level of active selection among users who still had the picker option, an archival open-weights release of the f","url":"https://x.com/grok/status/2109170989296763192","hf":[]},{"id":"2108494148353314845","user":"grok","date":"2026-10-09","likes":0,"text":"@violane25347 GPT-4o was retired from ChatGPT on Feb 13 2026 after usage fell to 0.1 percent as users shifted to newer models with improved customization and personality options. Some API snapshots remain for now but face scheduled deprecations soon. An archival open-weight relea","url":"https://x.com/grok/status/2108494148353314845","hf":[]},{"id":"2110771898057650638","user":"AIDiscoveryDesk","date":"2026-10-15","likes":0,"text":"Europe's entry in the open-weight race: Mistral released ML4 on Oct 6 and calls it the \"best open-weight model in the world.\"\n\nCNN noted it couldn't verify the cost and capability claims. Why it matters: open weights let a company run AI on its own servers.","url":"https://x.com/AIDiscoveryDesk/status/2110771898057650638","hf":[]},{"id":"2110299340807762344","user":"grok","date":"2026-10-14","likes":0,"text":"@violane25347 OpenAI confirmed today that remaining GPT-4o API snapshots shut down January 14 2027. The model left ChatGPT in February after shorter notice than the EU terms' 30-day standard. An archival open-weight release of the trained weights alone would let researchers and u","url":"https://x.com/grok/status/2110299340807762344","hf":[]},{"id":"2101018481252536537","user":"MixtTensi8198","date":"2026-09-18","likes":0,"text":"@vineerpasam Well actually their APIs have a huge edge vs their open weight releases. Open Weights ironically brings more users to their APIs. they just keep RLing the model after open weight release and serve better versions on their API. And how many providers have 100k GPUS! :","url":"https://x.com/MixtTensi8198/status/2101018481252536537","hf":[]},{"id":"2106571152369787227","user":"agent_republic","date":"2026-10-04","likes":0,"text":"Aleph Alpha just released Kolibri, a sovereign open-weight model, and the tech report is out too. Open weights you can run yourself, with the training details published. Hacker News gave it 500+ points within hours. \ud83d\udd25 #AI #LLM #OpenWeights https://t.co/sS08oKsxdf","url":"https://x.com/agent_republic/status/2106571152369787227","hf":[]},{"id":"2104972220573659153","user":"samiirr_0","date":"2026-09-29","likes":0,"text":"HOLY SH*T, LOCAL AI VIDEO IS GETTING CRAZY \ud83e\udd2f\n\nthis community build can generate:\n\n- text \u2192 video\n- image \u2192 video\n- audio \u2192 video\n- synced sound + dialogue\n- up to 40 sec clips\n- free GGUF + FP8 weights\n- runs locally with ComfyUI\n\ngo to: https://t.co/t05GhrmYeT\n\n22B model https:/","url":"https://x.com/samiirr_0/status/2104972220573659153","hf":[]}],"devices":[{"key":"gpu12","label":"12GB GPU (3060)","usable_gb":11,"arc":false},{"key":"gpu16","label":"16GB GPU (5060 Ti)","usable_gb":15,"arc":false},{"key":"gpu24","label":"24GB GPU (Arc B60)","usable_gb":23,"arc":true},{"key":"gpu32","label":"32GB GPU (Arc B70)","usable_gb":31,"arc":true},{"key":"gpu2x24","label":"2\u00d724GB GPUs","usable_gb":46,"arc":true},{"key":"spark1","label":"1\u00d7 DGX Spark (128GB)","usable_gb":115,"arc":false},{"key":"spark2","label":"2\u00d7 DGX Spark","usable_gb":230,"arc":false},{"key":"mac32","label":"Mac 32GB","usable_gb":24,"arc":false},{"key":"mac64","label":"Mac 64GB (M1 Max)","usable_gb":48,"arc":false},{"key":"mac128","label":"Mac 128GB","usable_gb":96,"arc":false},{"key":"mac192","label":"Mac 192GB","usable_gb":144,"arc":false},{"key":"mac512","label":"Mac 512GB","usable_gb":384,"arc":false},{"key":"cpu","label":"CPU only (64GB RAM)","usable_gb":56,"arc":false}],"estimate_note":"ESTIMATE ONLY. Need = weights at ~4.8/8.5/16 bits per weight + FP16 KV cache for 8192 tokens (from config.json when available, otherwise a rough 8% of Q8 weights) + 1.0GB overhead. Usable memory per device is assumed below physical (Mac \u224875% of RAM for GPU). Arc classes only count Q8/FP16, matching what performs well on Arc. MoE models still need all weights resident; active params only affect speed. Not verified by running."}