[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"$f342oj0u0wkev3":3,"$f38acka58proqr":73,"$f2dvo25sgd67uj":75,"$f3boz8zgtb23c3":426,"$f8ypa021er9qz":428,"$f2n9exo9fbhrd6":430},{"id":4,"name":5,"publisher":6,"paramsB":7,"useCase":8,"licenseStatus":9,"licenseUrl":10,"sourceUrl":11,"descriptionZh":12,"descriptionEn":13,"imageUrl":14,"downloadUrls":15,"stats":18,"variants":23,"derivations":24,"measuredHardware":27,"updatedAt":72},95,"elephant-1","Baseten (Base Labs)",9,"REASONING","OPEN_SOURCE","https:\u002F\u002Fopensource.org\u002Flicenses\u002FMIT","https:\u002F\u002Fhuggingface.co\u002Fbaselabs\u002Felephant-1","mammoth-1 的 9B 小兄弟（2026-10-21 发布，MIT，Qwen3.5-9B 全量微调），面向「快且便宜」的决策场景：同样单次 shared-prefix prefill 输出每个选项的校准概率。成绩：Jev Decision Index 0.4 上 15B 以下最佳（62.65，并列 208 个模型第 8）；Artificial Analysis Decision Index 复刻榜 65.2，与 d1-ultra（65）持平。与 mammoth-1 的分工：需要更广通用知识时用 27B，延迟 \u002F 成本敏感的路由与守门用 9B。自托管需官方 setup.sh + OneJev 运行时（单 CUDA GPU）。","The 9B sibling of mammoth-1 (released 2026-10-21, MIT, a full fine-tune of Qwen3.5-9B) for fast, cheap decisions: same single shared-prefix prefill returning calibrated probabilities for every option. Scores: best model under 15B on the Jev Decision Index 0.4 (62.65, tied #8 of 208) and 65.2 on a replica of the Artificial Analysis Decision Index, level with d1-ultra (65). Division of labor with mammoth-1: reach for the 27B when broader general knowledge matters, the 9B when latency and cost drive routing and guardrail duty. Self-hosting requires the official setup.sh with the OneJev runtime (a single CUDA GPU).",null,[16],{"url":11,"label":17},"HuggingFace",{"recordCount":19,"variantCount":20,"hardwareCount":21,"unlinkedRecordCount":19,"evidence":22},39,0,15,{"L0":20,"L1":20,"L2":19},[],{"parents":25,"children":26},[],[],[28,32,36,39,41,43,46,49,52,55,59,61,64,67,70],{"name":29,"vramGb":30,"recordCount":31},"NVIDIA RTX Spark",128,5,{"name":33,"vramGb":34,"recordCount":35},"NVIDIA RTX PRO 6000 Blackwell 96GB",96,4,{"name":37,"vramGb":30,"recordCount":38},"AMD Ryzen AI Max+ 395",3,{"name":40,"vramGb":14,"recordCount":38},"Apple M2 Ultra",{"name":42,"vramGb":30,"recordCount":38},"Apple M5 Max",{"name":44,"vramGb":45,"recordCount":38},"Apple M5 Ultra",512,{"name":47,"vramGb":48,"recordCount":38},"NVIDIA H20",97,{"name":50,"vramGb":51,"recordCount":38},"NVIDIA RTX 4090",24,{"name":53,"vramGb":54,"recordCount":38},"NVIDIA RTX 5090",32,{"name":56,"vramGb":57,"recordCount":58},"NVIDIA H100 80GB HBM3",80,2,{"name":60,"vramGb":51,"recordCount":58},"NVIDIA RTX 3090",{"name":62,"vramGb":63,"recordCount":58},"NVIDIA RTX 4080 Super",16,{"name":65,"vramGb":51,"recordCount":66},"AMD Radeon RX 7900 XTX",1,{"name":68,"vramGb":69,"recordCount":66},"NVIDIA RTX 3060 12GB",12,{"name":71,"vramGb":63,"recordCount":66},"NVIDIA RTX 5080","2026-10-22T15:13:14.232Z",{"favorited":74},false,{"items":76,"total":424,"page":66,"pageSize":425},[77,94,105,117,129,139,150,159,170,179,189,198,208,217,226,236,243,253,262,270,281,289,296,303,314,321,329,339,348,358,366,375,386,394,401,408,416],{"modelId":4,"modelName":5,"hardwareId":66,"hardwareName":50,"frameworkId":66,"frameworkName":78,"quantization":79,"hardwareVramGb":51,"decodeTps":80,"prefillTps":81,"vramGb":82,"genSeconds":14,"modelUseCase":8,"ttftS":83,"ttfbGb":84,"mtpAcceptanceRate":14,"powerW":85,"sampleCount":66,"reproduceCount":66,"sourceTypes":86,"evidenceLevel":88,"updatedAt":89,"latestFrameworkVersion":90,"latestOs":91,"latestCudaVersion":92,"latestFlashAttention":93,"latestOutputSpec":14},"llama.cpp","IQ1_S",180.81,654.72,10.04,0.191,1.45,395.6,[87],"OTHER","L2","2026-10-22T15:17:41.900Z","v0.46","Windows 11 23H2","12.8",true,{"modelId":4,"modelName":5,"hardwareId":66,"hardwareName":50,"frameworkId":66,"frameworkName":78,"quantization":95,"hardwareVramGb":51,"decodeTps":96,"prefillTps":97,"vramGb":98,"genSeconds":14,"modelUseCase":8,"ttftS":99,"ttfbGb":100,"mtpAcceptanceRate":14,"powerW":101,"sampleCount":66,"reproduceCount":66,"sourceTypes":102,"evidenceLevel":88,"updatedAt":89,"latestFrameworkVersion":103,"latestOs":91,"latestCudaVersion":104,"latestFlashAttention":93,"latestOutputSpec":14},"IQ3_XS",136.35,504.8,13.82,0.586,1.09,391.8,[87],"v0.29","12.4",{"modelId":4,"modelName":5,"hardwareId":66,"hardwareName":50,"frameworkId":58,"frameworkName":106,"quantization":107,"hardwareVramGb":51,"decodeTps":108,"prefillTps":109,"vramGb":110,"genSeconds":14,"modelUseCase":8,"ttftS":111,"ttfbGb":112,"mtpAcceptanceRate":14,"powerW":113,"sampleCount":66,"reproduceCount":66,"sourceTypes":114,"evidenceLevel":88,"updatedAt":89,"latestFrameworkVersion":115,"latestOs":116,"latestCudaVersion":104,"latestFlashAttention":93,"latestOutputSpec":14},"vLLM","IQ2_XS",179.71,772.31,11.17,1.047,1.44,394.8,[87],"v0.37","Ubuntu 24.04",{"modelId":4,"modelName":5,"hardwareId":58,"hardwareName":60,"frameworkId":38,"frameworkName":118,"quantization":119,"hardwareVramGb":51,"decodeTps":120,"prefillTps":121,"vramGb":122,"genSeconds":14,"modelUseCase":8,"ttftS":123,"ttfbGb":124,"mtpAcceptanceRate":14,"powerW":125,"sampleCount":66,"reproduceCount":66,"sourceTypes":126,"evidenceLevel":88,"updatedAt":89,"latestFrameworkVersion":127,"latestOs":91,"latestCudaVersion":128,"latestFlashAttention":93,"latestOutputSpec":14},"Ollama","Q4_K_S",69.84,337.36,16.84,0.565,0.56,251.1,[87],"v0.44","12.6",{"modelId":4,"modelName":5,"hardwareId":58,"hardwareName":60,"frameworkId":7,"frameworkName":130,"quantization":79,"hardwareVramGb":51,"decodeTps":131,"prefillTps":132,"vramGb":82,"genSeconds":14,"modelUseCase":8,"ttftS":133,"ttfbGb":134,"mtpAcceptanceRate":135,"powerW":136,"sampleCount":66,"reproduceCount":66,"sourceTypes":137,"evidenceLevel":88,"updatedAt":89,"latestFrameworkVersion":138,"latestOs":116,"latestCudaVersion":92,"latestFlashAttention":74,"latestOutputSpec":14},"Strata",163.87,748,1.438,1.31,0.6604,307.2,[87],"v0.26",{"modelId":4,"modelName":5,"hardwareId":140,"hardwareName":65,"frameworkId":58,"frameworkName":106,"quantization":141,"hardwareVramGb":51,"decodeTps":142,"prefillTps":143,"vramGb":144,"genSeconds":14,"modelUseCase":8,"ttftS":145,"ttfbGb":146,"mtpAcceptanceRate":14,"powerW":147,"sampleCount":66,"reproduceCount":66,"sourceTypes":148,"evidenceLevel":88,"updatedAt":89,"latestFrameworkVersion":149,"latestOs":91,"latestCudaVersion":92,"latestFlashAttention":93,"latestOutputSpec":14},11,"Q3_K_M",94.98,305.39,14.95,2.112,0.76,246.6,[87],"v0.32",{"modelId":4,"modelName":5,"hardwareId":151,"hardwareName":53,"frameworkId":66,"frameworkName":78,"quantization":119,"hardwareVramGb":54,"decodeTps":152,"prefillTps":153,"vramGb":122,"genSeconds":14,"modelUseCase":8,"ttftS":154,"ttfbGb":155,"mtpAcceptanceRate":14,"powerW":156,"sampleCount":66,"reproduceCount":66,"sourceTypes":157,"evidenceLevel":88,"updatedAt":89,"latestFrameworkVersion":158,"latestOs":116,"latestCudaVersion":92,"latestFlashAttention":93,"latestOutputSpec":14},13,219.01,766.77,0.604,1.75,353.5,[87],"v0.54",{"modelId":4,"modelName":5,"hardwareId":151,"hardwareName":53,"frameworkId":38,"frameworkName":118,"quantization":160,"hardwareVramGb":54,"decodeTps":161,"prefillTps":162,"vramGb":163,"genSeconds":14,"modelUseCase":8,"ttftS":164,"ttfbGb":165,"mtpAcceptanceRate":14,"powerW":166,"sampleCount":66,"reproduceCount":66,"sourceTypes":167,"evidenceLevel":88,"updatedAt":89,"latestFrameworkVersion":168,"latestOs":169,"latestCudaVersion":128,"latestFlashAttention":93,"latestOutputSpec":14},"Q5_K_S",174.79,860.94,19.11,0.233,1.4,378.4,[87],"v0.27","Windows 11 24H2",{"modelId":4,"modelName":5,"hardwareId":151,"hardwareName":53,"frameworkId":38,"frameworkName":118,"quantization":171,"hardwareVramGb":54,"decodeTps":172,"prefillTps":173,"vramGb":174,"genSeconds":14,"modelUseCase":8,"ttftS":175,"ttfbGb":176,"mtpAcceptanceRate":14,"powerW":177,"sampleCount":66,"reproduceCount":66,"sourceTypes":178,"evidenceLevel":88,"updatedAt":89,"latestFrameworkVersion":158,"latestOs":91,"latestCudaVersion":92,"latestFlashAttention":93,"latestOutputSpec":14},"Q8_0",138.97,570.96,23.84,0.285,1.11,379.5,[87],{"modelId":4,"modelName":5,"hardwareId":180,"hardwareName":71,"frameworkId":66,"frameworkName":78,"quantization":95,"hardwareVramGb":63,"decodeTps":181,"prefillTps":182,"vramGb":98,"genSeconds":14,"modelUseCase":8,"ttftS":183,"ttfbGb":184,"mtpAcceptanceRate":14,"powerW":185,"sampleCount":66,"reproduceCount":66,"sourceTypes":186,"evidenceLevel":88,"updatedAt":89,"latestFrameworkVersion":187,"latestOs":188,"latestCudaVersion":128,"latestFlashAttention":93,"latestOutputSpec":14},14,196.47,707.56,2.067,1.57,275.1,[87],"v0.31","Ubuntu 22.04",{"modelId":4,"modelName":5,"hardwareId":21,"hardwareName":68,"frameworkId":58,"frameworkName":106,"quantization":190,"hardwareVramGb":69,"decodeTps":191,"prefillTps":192,"vramGb":193,"genSeconds":14,"modelUseCase":8,"ttftS":194,"ttfbGb":124,"mtpAcceptanceRate":14,"powerW":195,"sampleCount":66,"reproduceCount":66,"sourceTypes":196,"evidenceLevel":88,"updatedAt":89,"latestFrameworkVersion":197,"latestOs":116,"latestCudaVersion":128,"latestFlashAttention":93,"latestOutputSpec":14},"Q2_K",70.07,259.24,11.4,0.621,147.9,[87],"v0.42",{"modelId":4,"modelName":5,"hardwareId":199,"hardwareName":62,"frameworkId":38,"frameworkName":118,"quantization":200,"hardwareVramGb":63,"decodeTps":201,"prefillTps":202,"vramGb":203,"genSeconds":14,"modelUseCase":8,"ttftS":204,"ttfbGb":205,"mtpAcceptanceRate":14,"powerW":206,"sampleCount":66,"reproduceCount":66,"sourceTypes":207,"evidenceLevel":88,"updatedAt":89,"latestFrameworkVersion":127,"latestOs":91,"latestCudaVersion":104,"latestFlashAttention":93,"latestOutputSpec":14},20,"Q3_K_L",145.8,516.93,15.2,1.048,1.17,292.4,[87],{"modelId":4,"modelName":5,"hardwareId":199,"hardwareName":62,"frameworkId":7,"frameworkName":130,"quantization":95,"hardwareVramGb":63,"decodeTps":209,"prefillTps":210,"vramGb":98,"genSeconds":14,"modelUseCase":8,"ttftS":211,"ttfbGb":212,"mtpAcceptanceRate":213,"powerW":214,"sampleCount":66,"reproduceCount":66,"sourceTypes":215,"evidenceLevel":88,"updatedAt":89,"latestFrameworkVersion":216,"latestOs":188,"latestCudaVersion":92,"latestFlashAttention":74,"latestOutputSpec":14},244.72,1076.88,1.663,1.96,0.4852,297.5,[87],"v0.30",{"modelId":4,"modelName":5,"hardwareId":51,"hardwareName":37,"frameworkId":66,"frameworkName":78,"quantization":200,"hardwareVramGb":30,"decodeTps":218,"prefillTps":219,"vramGb":220,"genSeconds":14,"modelUseCase":8,"ttftS":221,"ttfbGb":222,"mtpAcceptanceRate":14,"powerW":223,"sampleCount":66,"reproduceCount":66,"sourceTypes":224,"evidenceLevel":88,"updatedAt":89,"latestFrameworkVersion":225,"latestOs":91,"latestCudaVersion":104,"latestFlashAttention":93,"latestOutputSpec":14},32.72,137.17,15.71,2.45,0.26,52.8,[87],"v0.51",{"modelId":4,"modelName":5,"hardwareId":51,"hardwareName":37,"frameworkId":58,"frameworkName":106,"quantization":227,"hardwareVramGb":30,"decodeTps":228,"prefillTps":229,"vramGb":230,"genSeconds":14,"modelUseCase":8,"ttftS":231,"ttfbGb":232,"mtpAcceptanceRate":14,"powerW":233,"sampleCount":66,"reproduceCount":66,"sourceTypes":234,"evidenceLevel":88,"updatedAt":89,"latestFrameworkVersion":235,"latestOs":116,"latestCudaVersion":128,"latestFlashAttention":74,"latestOutputSpec":14},"Q6_K",22.65,118.27,21,1.87,0.18,63.5,[87],"v0.25",{"modelId":4,"modelName":5,"hardwareId":51,"hardwareName":37,"frameworkId":58,"frameworkName":106,"quantization":171,"hardwareVramGb":30,"decodeTps":237,"prefillTps":238,"vramGb":174,"genSeconds":14,"modelUseCase":8,"ttftS":239,"ttfbGb":240,"mtpAcceptanceRate":14,"powerW":241,"sampleCount":66,"reproduceCount":66,"sourceTypes":242,"evidenceLevel":88,"updatedAt":89,"latestFrameworkVersion":216,"latestOs":116,"latestCudaVersion":92,"latestFlashAttention":93,"latestOutputSpec":14},19.62,103.78,2.337,0.16,45.4,[87],{"modelId":4,"modelName":5,"hardwareId":244,"hardwareName":42,"frameworkId":66,"frameworkName":78,"quantization":160,"hardwareVramGb":30,"decodeTps":245,"prefillTps":246,"vramGb":163,"genSeconds":14,"modelUseCase":8,"ttftS":247,"ttfbGb":248,"mtpAcceptanceRate":14,"powerW":249,"sampleCount":66,"reproduceCount":66,"sourceTypes":250,"evidenceLevel":88,"updatedAt":89,"latestFrameworkVersion":251,"latestOs":252,"latestCudaVersion":14,"latestFlashAttention":93,"latestOutputSpec":14},28,59.51,259.7,1.229,0.48,59.9,[87],"v0.40","macOS 26.1",{"modelId":4,"modelName":5,"hardwareId":244,"hardwareName":42,"frameworkId":38,"frameworkName":118,"quantization":254,"hardwareVramGb":30,"decodeTps":255,"prefillTps":256,"vramGb":257,"genSeconds":14,"modelUseCase":8,"ttftS":258,"ttfbGb":259,"mtpAcceptanceRate":14,"powerW":260,"sampleCount":66,"reproduceCount":66,"sourceTypes":261,"evidenceLevel":88,"updatedAt":89,"latestFrameworkVersion":127,"latestOs":252,"latestCudaVersion":14,"latestFlashAttention":93,"latestOutputSpec":14},"FP16",30.75,124.45,37.06,0.389,0.25,65.2,[87],{"modelId":4,"modelName":5,"hardwareId":244,"hardwareName":42,"frameworkId":38,"frameworkName":118,"quantization":160,"hardwareVramGb":30,"decodeTps":263,"prefillTps":264,"vramGb":163,"genSeconds":14,"modelUseCase":8,"ttftS":265,"ttfbGb":266,"mtpAcceptanceRate":14,"powerW":267,"sampleCount":66,"reproduceCount":66,"sourceTypes":268,"evidenceLevel":88,"updatedAt":89,"latestFrameworkVersion":269,"latestOs":252,"latestCudaVersion":14,"latestFlashAttention":74,"latestOutputSpec":14},53.6,253.24,1.276,0.43,61.7,[87],"v0.39",{"modelId":4,"modelName":5,"hardwareId":271,"hardwareName":44,"frameworkId":66,"frameworkName":78,"quantization":272,"hardwareVramGb":45,"decodeTps":273,"prefillTps":274,"vramGb":275,"genSeconds":14,"modelUseCase":8,"ttftS":276,"ttfbGb":277,"mtpAcceptanceRate":14,"powerW":278,"sampleCount":66,"reproduceCount":66,"sourceTypes":279,"evidenceLevel":88,"updatedAt":89,"latestFrameworkVersion":280,"latestOs":252,"latestCudaVersion":14,"latestFlashAttention":93,"latestOutputSpec":14},29,"Q3_K_S",117.38,386.02,14.38,2.43,0.94,125.4,[87],"v0.34",{"modelId":4,"modelName":5,"hardwareId":271,"hardwareName":44,"frameworkId":66,"frameworkName":78,"quantization":160,"hardwareVramGb":45,"decodeTps":282,"prefillTps":283,"vramGb":163,"genSeconds":14,"modelUseCase":8,"ttftS":284,"ttfbGb":285,"mtpAcceptanceRate":14,"powerW":286,"sampleCount":66,"reproduceCount":66,"sourceTypes":287,"evidenceLevel":88,"updatedAt":89,"latestFrameworkVersion":288,"latestOs":252,"latestCudaVersion":14,"latestFlashAttention":93,"latestOutputSpec":14},83.32,309.36,0.214,0.67,127.9,[87],"v0.52",{"modelId":4,"modelName":5,"hardwareId":271,"hardwareName":44,"frameworkId":38,"frameworkName":118,"quantization":227,"hardwareVramGb":45,"decodeTps":290,"prefillTps":291,"vramGb":230,"genSeconds":14,"modelUseCase":8,"ttftS":292,"ttfbGb":293,"mtpAcceptanceRate":14,"powerW":294,"sampleCount":66,"reproduceCount":66,"sourceTypes":295,"evidenceLevel":88,"updatedAt":89,"latestFrameworkVersion":127,"latestOs":252,"latestCudaVersion":14,"latestFlashAttention":74,"latestOutputSpec":14},62.32,341.91,1.703,0.5,130,[87],{"modelId":4,"modelName":5,"hardwareId":297,"hardwareName":56,"frameworkId":38,"frameworkName":118,"quantization":141,"hardwareVramGb":57,"decodeTps":298,"prefillTps":299,"vramGb":144,"genSeconds":14,"modelUseCase":8,"ttftS":300,"ttfbGb":276,"mtpAcceptanceRate":14,"powerW":301,"sampleCount":66,"reproduceCount":66,"sourceTypes":302,"evidenceLevel":88,"updatedAt":89,"latestFrameworkVersion":158,"latestOs":116,"latestCudaVersion":92,"latestFlashAttention":93,"latestOutputSpec":14},55,303.51,1532.77,2.332,621.3,[87],{"modelId":4,"modelName":5,"hardwareId":297,"hardwareName":56,"frameworkId":7,"frameworkName":130,"quantization":304,"hardwareVramGb":57,"decodeTps":305,"prefillTps":306,"vramGb":307,"genSeconds":14,"modelUseCase":8,"ttftS":308,"ttfbGb":309,"mtpAcceptanceRate":310,"powerW":311,"sampleCount":66,"reproduceCount":66,"sourceTypes":312,"evidenceLevel":88,"updatedAt":89,"latestFrameworkVersion":115,"latestOs":188,"latestCudaVersion":313,"latestFlashAttention":74,"latestOutputSpec":14},"IQ3_S",322.41,1596.98,14.57,0.396,2.58,0.774,587.5,[87],"13.0",{"modelId":4,"modelName":5,"hardwareId":315,"hardwareName":47,"frameworkId":58,"frameworkName":106,"quantization":254,"hardwareVramGb":48,"decodeTps":316,"prefillTps":317,"vramGb":257,"genSeconds":14,"modelUseCase":8,"ttftS":293,"ttfbGb":318,"mtpAcceptanceRate":14,"powerW":319,"sampleCount":66,"reproduceCount":66,"sourceTypes":320,"evidenceLevel":88,"updatedAt":89,"latestFrameworkVersion":269,"latestOs":169,"latestCudaVersion":92,"latestFlashAttention":74,"latestOutputSpec":14},56,25.76,119.08,0.21,304.2,[87],{"modelId":4,"modelName":5,"hardwareId":315,"hardwareName":47,"frameworkId":58,"frameworkName":106,"quantization":107,"hardwareVramGb":48,"decodeTps":322,"prefillTps":323,"vramGb":110,"genSeconds":14,"modelUseCase":8,"ttftS":324,"ttfbGb":325,"mtpAcceptanceRate":14,"powerW":326,"sampleCount":66,"reproduceCount":66,"sourceTypes":327,"evidenceLevel":88,"updatedAt":89,"latestFrameworkVersion":328,"latestOs":188,"latestCudaVersion":128,"latestFlashAttention":93,"latestOutputSpec":14},68.56,374.49,2.172,0.55,384.5,[87],"v0.49",{"modelId":4,"modelName":5,"hardwareId":315,"hardwareName":47,"frameworkId":38,"frameworkName":118,"quantization":330,"hardwareVramGb":48,"decodeTps":331,"prefillTps":332,"vramGb":333,"genSeconds":14,"modelUseCase":8,"ttftS":334,"ttfbGb":335,"mtpAcceptanceRate":14,"powerW":336,"sampleCount":66,"reproduceCount":66,"sourceTypes":337,"evidenceLevel":88,"updatedAt":89,"latestFrameworkVersion":338,"latestOs":169,"latestCudaVersion":104,"latestFlashAttention":93,"latestOutputSpec":14},"Q5_K_M",36.62,154.85,19.68,1.315,0.29,354.1,[87],"v0.48",{"modelId":4,"modelName":5,"hardwareId":340,"hardwareName":33,"frameworkId":38,"frameworkName":118,"quantization":79,"hardwareVramGb":34,"decodeTps":341,"prefillTps":342,"vramGb":82,"genSeconds":14,"modelUseCase":8,"ttftS":343,"ttfbGb":344,"mtpAcceptanceRate":14,"powerW":345,"sampleCount":66,"reproduceCount":66,"sourceTypes":346,"evidenceLevel":88,"updatedAt":89,"latestFrameworkVersion":347,"latestOs":91,"latestCudaVersion":92,"latestFlashAttention":93,"latestOutputSpec":14},68,275.81,1195.7,2.256,2.21,333.6,[87],"v0.28",{"modelId":4,"modelName":5,"hardwareId":340,"hardwareName":33,"frameworkId":38,"frameworkName":118,"quantization":349,"hardwareVramGb":34,"decodeTps":350,"prefillTps":351,"vramGb":352,"genSeconds":14,"modelUseCase":8,"ttftS":353,"ttfbGb":354,"mtpAcceptanceRate":14,"powerW":355,"sampleCount":66,"reproduceCount":66,"sourceTypes":356,"evidenceLevel":88,"updatedAt":89,"latestFrameworkVersion":357,"latestOs":116,"latestCudaVersion":128,"latestFlashAttention":93,"latestOutputSpec":14},"IQ4_XS",194.74,889.56,16.09,2.046,1.56,372.8,[87],"v0.24",{"modelId":4,"modelName":5,"hardwareId":340,"hardwareName":33,"frameworkId":7,"frameworkName":130,"quantization":95,"hardwareVramGb":34,"decodeTps":359,"prefillTps":360,"vramGb":98,"genSeconds":14,"modelUseCase":8,"ttftS":361,"ttfbGb":362,"mtpAcceptanceRate":363,"powerW":364,"sampleCount":66,"reproduceCount":66,"sourceTypes":365,"evidenceLevel":88,"updatedAt":89,"latestFrameworkVersion":138,"latestOs":169,"latestCudaVersion":104,"latestFlashAttention":93,"latestOutputSpec":14},354.49,1856.14,0.886,2.84,0.6014,332.5,[87],{"modelId":4,"modelName":5,"hardwareId":340,"hardwareName":33,"frameworkId":7,"frameworkName":130,"quantization":119,"hardwareVramGb":34,"decodeTps":367,"prefillTps":368,"vramGb":122,"genSeconds":14,"modelUseCase":8,"ttftS":369,"ttfbGb":370,"mtpAcceptanceRate":371,"powerW":372,"sampleCount":66,"reproduceCount":66,"sourceTypes":373,"evidenceLevel":88,"updatedAt":89,"latestFrameworkVersion":374,"latestOs":116,"latestCudaVersion":104,"latestFlashAttention":93,"latestOutputSpec":14},293.11,1298.77,0.975,2.34,0.5414,351.4,[87],"v0.22",{"modelId":4,"modelName":5,"hardwareId":376,"hardwareName":40,"frameworkId":66,"frameworkName":78,"quantization":377,"hardwareVramGb":14,"decodeTps":378,"prefillTps":379,"vramGb":380,"genSeconds":14,"modelUseCase":8,"ttftS":381,"ttfbGb":382,"mtpAcceptanceRate":14,"powerW":383,"sampleCount":66,"reproduceCount":66,"sourceTypes":384,"evidenceLevel":88,"updatedAt":89,"latestFrameworkVersion":385,"latestOs":252,"latestCudaVersion":14,"latestFlashAttention":74,"latestOutputSpec":14},103,"Q4_K_M",92.52,377.54,17.6,2.376,0.74,110.5,[87],"v0.38",{"modelId":4,"modelName":5,"hardwareId":376,"hardwareName":40,"frameworkId":66,"frameworkName":78,"quantization":119,"hardwareVramGb":14,"decodeTps":387,"prefillTps":388,"vramGb":122,"genSeconds":14,"modelUseCase":8,"ttftS":389,"ttfbGb":390,"mtpAcceptanceRate":14,"powerW":391,"sampleCount":66,"reproduceCount":66,"sourceTypes":392,"evidenceLevel":88,"updatedAt":89,"latestFrameworkVersion":393,"latestOs":252,"latestCudaVersion":14,"latestFlashAttention":93,"latestOutputSpec":14},88.6,484.39,1.135,0.71,121.5,[87],"v0.35",{"modelId":4,"modelName":5,"hardwareId":376,"hardwareName":40,"frameworkId":38,"frameworkName":118,"quantization":227,"hardwareVramGb":14,"decodeTps":395,"prefillTps":396,"vramGb":230,"genSeconds":14,"modelUseCase":8,"ttftS":397,"ttfbGb":398,"mtpAcceptanceRate":14,"powerW":399,"sampleCount":66,"reproduceCount":66,"sourceTypes":400,"evidenceLevel":88,"updatedAt":89,"latestFrameworkVersion":158,"latestOs":252,"latestCudaVersion":14,"latestFlashAttention":93,"latestOutputSpec":14},67.84,354.32,1.451,0.54,100.3,[87],{"modelId":4,"modelName":5,"hardwareId":402,"hardwareName":29,"frameworkId":66,"frameworkName":78,"quantization":349,"hardwareVramGb":30,"decodeTps":403,"prefillTps":404,"vramGb":352,"genSeconds":14,"modelUseCase":8,"ttftS":405,"ttfbGb":266,"mtpAcceptanceRate":14,"powerW":406,"sampleCount":66,"reproduceCount":66,"sourceTypes":407,"evidenceLevel":88,"updatedAt":89,"latestFrameworkVersion":328,"latestOs":116,"latestCudaVersion":313,"latestFlashAttention":93,"latestOutputSpec":14},110,53.68,258.94,1.007,78.2,[87],{"modelId":4,"modelName":5,"hardwareId":402,"hardwareName":29,"frameworkId":66,"frameworkName":78,"quantization":377,"hardwareVramGb":30,"decodeTps":409,"prefillTps":410,"vramGb":380,"genSeconds":14,"modelUseCase":8,"ttftS":411,"ttfbGb":412,"mtpAcceptanceRate":14,"powerW":413,"sampleCount":58,"reproduceCount":66,"sourceTypes":414,"evidenceLevel":88,"updatedAt":89,"latestFrameworkVersion":415,"latestOs":116,"latestCudaVersion":313,"latestFlashAttention":93,"latestOutputSpec":14},50.55,236.22,1.013,0.41,54.05,[87],"v0.50",{"modelId":4,"modelName":5,"hardwareId":402,"hardwareName":29,"frameworkId":38,"frameworkName":118,"quantization":377,"hardwareVramGb":30,"decodeTps":417,"prefillTps":418,"vramGb":380,"genSeconds":14,"modelUseCase":8,"ttftS":419,"ttfbGb":420,"mtpAcceptanceRate":14,"powerW":421,"sampleCount":58,"reproduceCount":66,"sourceTypes":422,"evidenceLevel":88,"updatedAt":89,"latestFrameworkVersion":423,"latestOs":116,"latestCudaVersion":313,"latestFlashAttention":93,"latestOutputSpec":14},49.53,221.38,0.895,0.4,60.3,[87],"v0.41",37,50,{"items":427,"total":20,"page":66,"pageSize":199},[],{"items":429},[],{"items":431,"total":66,"page":66,"pageSize":38},[432],{"slug":433,"category":434,"publishedAt":435,"titleZh":436,"titleEn":437,"summaryZh":438,"summaryEn":439,"models":440,"hardwares":450},"krafton-raon-1-mini-2026-10-22","NEWS","2026-10-22T14:00:00.000Z","KRAFTON Raon-1-mini 预览版：28.8B-A4.3B 混合注意力 MoE，NVFP4 单卡 5090 可跑","KRAFTON Raon-1-mini preview: a 28.8B-A4.3B hybrid-attention MoE that fits a single RTX 5090 at NVFP4","10 月 22 日 KRAFTON AI 公开 Raon-1-mini 预览版（Apache-2.0，申请制自动批准）——从零训练的韩语强项 MoE：28.8B 总参 \u002F 4.3B 激活，25 层里 18 层 Raon Delta Attention（线性注意力）+ 7 层 Gated MLA（KV 约 8KiB\u002Ftoken），262K 上下文。官方 NVFP4 单卡 RTX 5090 可推理，DFlash2 草稿模型 B300 实测最高 2.18 倍吞吐。官方 GGUF Q4_K_M 19.7GB。站内新增 24 条实测。","KRAFTON AI published the Raon-1-mini preview on Oct 22 (Apache-2.0, request-to-access with automatic approval) — a from-scratch MoE strong at Korean: 28.8B total \u002F 4.3B active, with 18 of its 25 layers using Raon Delta Attention (linear attention) and 7 using Gated MLA (KV of ~8KiB per token), 262K context. Official NVFP4 runs on a single RTX 5090; the DFlash2 draft model delivers up to 2.18x throughput on a B300. Official GGUF Q4_K_M is 19.7GB. 24 new on-site runs.",[441,443,446,449],{"id":31,"name":442},"Qwen3.8-27B",{"id":444,"name":445},47,"Qwen3.6-35B-A3B",{"id":447,"name":448},93,"Raon-1-mini-preview",{"id":4,"name":5},[451,452],{"id":151,"name":53},{"id":402,"name":29}]