{"@context":"https://schema.org","@type":"NewsArticle","generatedAt":"2026-07-23T06:40:50.084Z","headline":"德国AI联盟发布开源大模型Soofi S 30B-A3B，基准测试领先","description":"德国AI协会协调的研究联盟发布开源大语言模型Soofi S，总参数316亿，每token仅激活约32亿参数。该模型采用混合Mamba-Transformer架构，在德语和英语基准测试中超越OLMo 3 32B和Apertus 70B等完全开源模型。","url":"https://www.aioga.com/news/cmro1333q010ebij2087uwrac/","mainEntityOfPage":"https://www.aioga.com/news/cmro1333q010ebij2087uwrac/","datePublished":"2026-07-16T21:28:29.737Z","dateModified":"2026-07-16T21:28:29.737Z","inLanguage":"zh-CN","publisher":{"@type":"NewsMediaOrganization","name":"Aioga","url":"https://www.aioga.com"},"citation":["https://the-decoder.com/german-ai-consortium-releases-soofi-s-an-open-30b-model-that-tops-benchmarks-in-both-english-and-german","https://aihot.virxact.com/items/cmro1333q010ebij2087uwrac"],"canonicalUrl":"https://www.aioga.com/news/cmro1333q010ebij2087uwrac/","directAnswer":{"@type":"Answer","text":"Aioga 编辑摘要：德国AI协会协调的研究联盟发布开源大语言模型Soofi S，总参数316亿，每token仅激活约32亿参数。 Aioga 将其归入「模型更新」方向，重点关注它对真实使用和行业竞争的影响。","url":"https://www.aioga.com/news/cmro1333q010ebij2087uwrac/","dateCreated":"2026-07-16T21:28:29.737Z","author":{"@type":"Organization","@id":"https://www.aioga.com/authors/aioga-editorial/#editorial-team","name":"Aioga Editorial Team","url":"https://www.aioga.com/authors/aioga-editorial/"}},"evidence":[{"@type":"CreativeWork","name":"the-decoder.com source article","url":"https://the-decoder.com/german-ai-consortium-releases-soofi-s-an-open-30b-model-that-tops-benchmarks-in-both-english-and-german","datePublished":"2026-07-16T21:28:29.737Z","provider":{"@type":"Organization","name":"the-decoder.com","url":"https://the-decoder.com/german-ai-consortium-releases-soofi-s-an-open-30b-model-that-tops-benchmarks-in-both-english-and-german"}},{"@type":"CreativeWork","name":"AIHot archive record","url":"https://aihot.virxact.com/items/cmro1333q010ebij2087uwrac","datePublished":"2026-07-16T21:28:29.737Z","provider":{"@type":"Organization","name":"AIHot","url":"https://aihot.virxact.com/items/cmro1333q010ebij2087uwrac"}}],"aggregationSource":"Hacker News 热门（buzzing.cc 中文翻译）","originalPublisher":{"name":"the-decoder.com","url":"https://the-decoder.com/german-ai-consortium-releases-soofi-s-an-open-30b-model-that-tops-benchmarks-in-both-english-and-german"},"article":{"id":"cmro1333q010ebij2087uwrac","slug":"cmro1333q010ebij2087uwrac","url":"https://www.aioga.com/news/cmro1333q010ebij2087uwrac/","title":"德国AI联盟发布开源大模型Soofi S 30B-A3B，基准测试领先","title_en":"德国人工智能联盟发布了Soofi S--一款位居基准测试榜首的300亿参数开源模型","summary":"德国AI协会协调的研究联盟发布开源大语言模型Soofi S，总参数316亿，每token仅激活约32亿参数。该模型采用混合Mamba-Transformer架构，在德语和英语基准测试中超越OLMo 3 32B和Apertus 70B等完全开源模型。","source":"Hacker News 热门（buzzing.cc 中文翻译）","sourceUrl":"https://the-decoder.com/german-ai-consortium-releases-soofi-s-an-open-30b-model-that-tops-benchmarks-in-both-english-and-german","aiHotUrl":"https://aihot.virxact.com/items/cmro1333q010ebij2087uwrac","publishedAt":"2026-07-16T21:28:29.737Z","category":"模型更新","score":57,"selected":false,"articleBody":["Soofi S is one of the first large language models trained entirely on Deutsche Telekom's Industrial AI Cloud in Munich. The open 30B model uses a lean hybrid architecture and a training mix deliberately weighted toward German.","Soofi S is a mixture-of-experts model. It contains 31.6 billion parameters in total but activates only about 3.2 billion per generated token. That puts its compute cost closer to a 3B model than a conventional 30B model. The consortium adopts the architecture of Nvidia's Nemotron 3 Nano：https://the-decoder.com/nvidias-nemotron-3-ultra-becomes-the-smartest-open-us-model-but-china-still-leads/ without modification, a hybrid design combining Mamba-2 layers with standard attention layers. Ad","The key difference from typical transformers is memory behavior. In conventional models, the KV cache that stores previous tokens for attention computation grows linearly with context length. With long inputs and many parallel requests, reloading that cache becomes a bottleneck. Only 6 of Soofi S's 52 layers maintain such a cache at all. Ad DEC_D_Incontent-1","The practical payoff shows up in generation throughput. At a context length of 40,000 tokens with 32 parallel requests, Soofi S generates roughly eight times more tokens per second per GPU than dense models in the 14 to 24 billion parameter range. While throughput drops significantly for conventional models as context grows, Soofi S stays nearly flat from 4,000 to 256,000 tokens. The only model that shows similar behavior in the measurements is Alibaba's Qwen3.5 35B-A3B：https://the-decoder.com/alibabas-open-qwen-3-5-takes-aim-at-gpt-5-mini-and-claude-sonnet-4-5-at-a-fraction-of-the-cost/, which also uses a hybrid architecture.","The consortium processed about 27 trillion tokens in total, split across three phases. In the first phase, the model learns language fundamentals from roughly 20 trillion tokens drawn from a broad mix of web, code, math, and domain-specific texts. A second phase follows with about 6 trillion tokens from higher-quality sources, designed to sharpen the patterns learned earlier. A shorter third phase then extends the context window by training on very long documents of up to one million tokens. Ad","The deliberate focus on German is central. In the first phase, German makes up 7.2 percent of the training mix; in the second phase, that share rises to 15.3 percent. In Nvidia's Nemotron reference recipe, all non-English languages combined account for only about 5 percent.","For data sources, the consortium combines German web text from HPLT, the openly licensed German Commons corpus, German portions of FinePDFs and FineWiki, and the commercially licensed Genios corpus containing 193 million newspaper articles from 916 German publications. Machine-translated and synthetically generated German texts round out the mix. Ad DEC_D_Incontent-2","In evaluations against 16 other open models, Soofi S leads all fully open models on aggregate scores for both German and English, according to the report. That includes OLMo 3 32B from the Allen Institute for AI and Apertus 70B from ETH Zurich and EPFL. Against every European sovereign baseline, the model comes out ahead on all German benchmarks in the suite, sometimes by double-digit margins. Ad","On code benchmarks, Soofi S scores 73.8 percent on HumanEval, 70.2 on MBPP, and 84.2 on the German MBPP variant, the best results among open-source peers. On INCLUDE-DE, a test for Germany-specific regional knowledge, Soofi S ties for first place at 61.2 points with the larger Qwen3.5 35B-A3B. Compared to the Nemotron baseline, the German data recipe improves language proficiency by 15.1 points and the science test GPQA-Diamond by 9.6 points, without sacrificing English performance.","Soofi S doesn't do as well on German competition math, where it scores 56 points on Minerva MATH-DE, well behind Qwen3.5 35B-A3B (76.5) and Gemma 3 27B：https://the-decoder.com/google-releases-new-gemma-3-open-model-family/ (65.6). It also lags on open factual retrieval in NaturalQuestions. The latter likely relates to having only 3 billion active parameters, which can store less world knowledge：https://the-decoder.com/sinas-open-model-vibethinker-3b-aims-to-show-reasoning-compresses-well-but-factual-knowledge-doesnt/ than a dense 27B model.","The RULER long-context test also reveals a specific weakness: When the model has to extract frequently occurring words from a long text, Soofi S's hit rate drops to around 3 percent beyond 32,000 tokens of context, while the comparable Nemotron model still manages 60 to 64 percent. The authors attribute this to the fact that their long-context training data contains many long documents but lacks synthetic data designed for extraction tasks. On the remaining twelve RULER tasks, both models perform about the same.","The training run took place between March and May on up to 512 Nvidia B200 GPUs at Deutsche Telekom's Industrial AI Cloud：https://the-decoder.com/10000-nvidia-blackwell-gpus-set-to-increase-germanys-ai-capacity-by-50-percent/ in Munich, totaling about 253,000 GPU-hours. According to the report, the facility runs entirely on renewable energy, is cooled with water from the Eisbach canal, and feeds waste heat into the surrounding Tucherpark neighborhood. Soofi S was one of the first major training runs on this infrastructure.","Behind Soofi is a consortium of German research institutions and companies, coordinated by the German AI Association and funded by the German Federal Ministry for Economic Affairs and Energy as part of the European IPCEI-CIS program.","Participants include the Fraunhofer Institutes IAIS and IIS, the German Research Center for Artificial Intelligence (DFKI), TU Darmstadt, the University of Würzburg, the L3S Research Center, the Berlin University of Applied Sciences, and AI companies Ellamind and Merantix Momentum. The project's goal is to build an open European AI model family that can run on sovereign infrastructure and be tested in industrial applications.","The researchers are releasing model weights along with selected intermediate checkpoints：https://huggingface.co/collections/Soofi-Project/soofi-s-beta-models, the complete training and evaluation code, and a detailed data inventory listing raw token counts, epoch numbers, and effective contributions per source. Sources that were reviewed but excluded are also documented. According to the team, this means Soofi S meets the Open Source AI Definition 1.0 from the Open Source Initiative：https://the-decoder.com/open-source-initiative-releases-first-formal-definition-of-open-source-ai/.","A stricter proposal for a European open-data definition, which would require every single training token to be freely distributable, isn't met because of the 1.3 percent share of Genios data, which carries a commercial license. The report says about 99 percent of the training mix can be independently reconstructed. The exact license for the model's release hasn't been finalized yet.","As lead author Michael Fromm writes：https://x.com/effi288/status/2075904321707798699, Soofi S positions itself between broadly multilingual European sovereignty projects like EuroLLM or Teuken：https://the-decoder.com/eu-project-releases-7b-model-that-speaks-24-european-languages/, which cover many languages, and the highest-performing international open-weight models. According to the project website, the consortium is looking for industry partners for the next phase to test the model in applications involving technical documents, code generation, and agent-based systems.","Stay in the loop on AI. Clear, useful, no fluff.","Follow The Decoder for AI news, background stories and expert analyses.","The Decoder：https://the-decoder.com/"],"articleImages":[{"sourceUrl":"https://the-decoder.com/wp-content/uploads/2026/07/soofi-s-03-data-mixture.jpg","alt":"Flow chart showing the training data mix across three phases. Seven categories including English Web, Code, Reasoning, Math, and German shift from Phase 1 (about 23T tokens) through Phase 2 (about 6T) to Phase 3 (about 188B), with the German share rising from 7.2 to 15.3 percent.","afterParagraph":4,"url":"/media/articles/cmrj6actv0651bilkm5pfz6ub/bc1678a2d239b8cb.jpg"}],"mediaStatus":"ok","articleBodyZh":["Soofi S是首批完全在慕尼黑德国电信工业人工智能云上训练的大型语言模型之一。OPEN 30B模型采用精益混合架构和故意向德国加权的训练组合。","Soofi S是一个专家混合模型。它总共包含316亿个参数，但每个生成的代币仅激活约32亿个。这使得它的计算成本比传统的30B模型更接近3B模型。 联合体采用英伟达Nemotron 3 Nano架构： https://the-decoder.com/nvidias-nemotron-3-ultra-becomes-the-smartest-open-us-model-but-china-still-leads/未经修改，采用Mamba-2层与标准关注层相结合的混合设计。广告","与典型变压器的关键区别在于记忆行为。在传统模型中，存储用于注意力计算的先前令牌的KV缓存随上下文长度线性增长。对于长输入和许多并行请求，重新加载缓存成为瓶颈。Soofi S的52层中只有6层保留了这样的缓存。广告DEC_D_Incontent-1","实际回报体现在发电吞吐量上。Soofi S的上下文长度为40,000个令牌， 32个并行请求，每个GPU每秒生成的令牌大约是140亿到240亿个参数范围内密集模型的8倍。虽然随着环境的增长，传统模型的吞吐量显着下降，但Soofi S的吞吐量几乎保持不变，从4,000个代币降至256,000个代币。 在测量中显示类似行为的唯一模型是阿里巴巴的Qwen3.5 35B-A3B： https://the-decoder.com/alibabas-open-qwen-3-5-takes-aim-at-gpt-5-mini-and-claude-sonnet-4-5-at-a-fraction-of-the-cost/，它也使用混合架构。","该联盟总共处理了大约27万亿个代币，分为三个阶段。在第一阶段，该模型从大约20万亿个令牌中学习语言基础知识，这些令牌来自网络、代码、数学和特定领域的广泛混合文本。第二阶段接下来是来自更高质量来源的大约6万亿个代币，旨在锐化之前学习的模式。 然后，较短的第三阶段通过对多达一百万个令牌的非常长的文档进行培训来扩展上下文窗口。广告","对德语的刻意关注是核心。在第一阶段，德语占培训组合的7.2%；在第二阶段，这一比例上升至15.3%。在Nvidia的Nemotron参考配方中，所有非英语语言合计仅占约5%。","对于数据源，该联盟结合了来自HPLT的德语Web文本、公开许可的德国公共语料库、FinePDF和FineWiki的德语部分以及商业许可的Genios语料库，其中包含来自916家德国出版物的1.93亿篇报纸文章。机器翻译和合成生成的德语文本完美融合。广告DEC_D_Incontent-2","根据该报告，在对其他16个开放模型的评估中， Soofi S在德语和英语的总分上领先于所有完全开放的模型。其中包括来自艾伦人工智能研究所的OLMo 3 32B和来自苏黎世联邦理工学院和EPFL的Apertus 70B。相对于每个欧洲主权基准，该模型在套件中的所有德国基准上都领先，有时以两位数的利润率领先。广告","在代码基准测试中， Soofi S在HumanEval上的得分为73.8%，在MBPP上的得分为70.2，在德国MBPP变体上的得分为84.2，这是开源同行中最好的结果。在针对德国特定区域知识的INCLUDE-DE测试中， Soofi S以61.2分与较大的Qwen3.5 35B-A3B并列第一。 与Nemotron基线相比，德国数据配方将语言熟练度提高了15.1分，科学测试GPQA-Diamond提高了9.6分，而不会牺牲英语表现。","Soofi S在德国竞争数学方面表现不佳，在Minerva MATH-DE上得分56分，远远落后于Qwen3.5 35B-A3B （ 76.5 ）和Gemma 3 27B： https://the-decoder.com/google-releases-new-gemma-3-open-model-family/ （ 65.6 ）。它在NaturalQuestions中的开放式事实检索方面也滞后。 后者可能与只有30亿个活动参数有关，与密集的27B模型相比，它可以存储更少的世界知识： https://the-decoder.com/sinas-open-model-vibethinker-3b-aims-to-show-reasoning-compresses-well-but-factual-knowledge-doesnt/。","标尺长上下文测试还揭示了一个特定的弱点：当模型必须从长文本中提取频繁出现的单词时， Soofi S的命中率在32,000个上下文令牌之外下降到约3 ％，而可比较的Nemotron模型仍然管理着60 ％至64 ％。 作者将此归因于他们的长上下文训练数据包含许多长文档，但缺乏专为提取任务设计的合成数据。在其余的十二个标尺任务中，两个模型的执行情况大致相同。","培训于3月至5月期间在德国电信位于慕尼黑的工业AI云（ https://the-decoder.com/10000-nvidia-blackwell-gpus-set-to-increase-germanys-ai-capacity-by-50-percent/ ）上使用多达512个Nvidia B200 GPU进行，总计约253,000个GPU小时。 根据该报告，该设施完全依靠可再生能源运行，用Eisbach运河的水冷却，并向周围的Tucherpark社区供给废热。Soofi S是这一基础设施的首批主要培训项目之一。","Soofi背后是一个由德国研究机构和公司组成的联盟，由德国人工智能协会协调，由德国联邦经济事务和能源部资助，作为欧洲IPCEI-CIS计划的一部分。","参与者包括弗劳恩霍夫研究所IAIS和IIS、德国人工智能研究中心（ DFKI ）、达姆施塔特工业大学、维尔茨堡大学、L3S研究中心、柏林应用科学大学以及人工智能公司Ellamind和Merantix Momentum。该项目的目标是建立一个开放的欧洲人工智能模型家族，可以在主权基础设施上运行，并在工业应用中进行测试。","研究人员正在发布模型权重以及选定的中间检查点： https://huggingface.co/collections/Soofi-Project/soofi-s-beta-models、完整的训练和评估代码以及详细的数据清单，其中列出了原始令牌计数、纪元编号和每个来源的有效贡献。还记录了已审核但排除的来源。 根据该团队的说法，这意味着Soofi S符合开源倡议（ Open Source Initiative ）的开源AI定义1.0： https://the-decoder.com/open-source-initiative-releases-first-formal-definition-of-open-source-ai/。","一项更严格的欧洲开放数据定义提案要求每一个培训令牌都是可自由分发的，但由于Genios数据的1.3%份额（带有商业许可证）而没有得到满足。报告称，大约99%的训练组合可以独立重建。该模型发布的确切许可证尚未最终确定。","正如主要作者Michael Fromm所写： https://x.com/effi288/status/2075904321707798699， Soofi S将自己定位在广泛的多语言欧洲主权项目之间，如EuroLLM或Teuken： https://the-decoder.com/eu-project-releases-7b-model-that-speaks-24-european-languages/，涵盖多种语言，以及性能最高的国际开放权重模型。 根据该项目网站，该联盟正在寻找下一阶段的行业合作伙伴，以便在涉及技术文档、代码生成和基于代理的系统的应用程序中测试该模型。","随时掌握人工智能的最新动态。清晰、实用、无绒毛。","关注解码器，了解AI新闻、背景故事和专家分析。","解码器： https://the-decoder.com/"],"translationStatus":"translated","bodyOrigin":"source-page","editorial":{"summary":"Aioga 编辑摘要：德国AI协会协调的研究联盟发布开源大语言模型Soofi S，总参数316亿，每token仅激活约32亿参数。 Aioga 将其归入「模型更新」方向，重点关注它对真实使用和行业竞争的影响。","background":"背景分析：模型与研究类动态需要结合能力边界、开放方式、成本、可用性和真实任务表现判断，单项指标领先不等于已经形成稳定采用。","viewpoint":"Aioga 判断：这条动态更适合作为行业观察信号，当前信息足以建立线索，但不足以推导长期结论。","implications":"影响分析：对相关团队而言，短期应先核对来源、可用范围和实际成本，再判断是否值得接入或跟进。","nextStep":"后续观察：继续观察官方文档、实际可用性、价格变化、开发者反馈和竞品回应。","evidenceRefs":["title","summary","articleBody"],"confidence":"medium","status":"published","aiGenerated":false,"autoApproved":true,"generatedBy":"rule-safe-fallback","generatedAt":"2026-07-23T06:49:19.197Z","sourceHash":"486ca3f1cfe562ff","validation":{"passed":true,"mode":"rule-safe-fallback","checks":["schema","length","source-attribution","no-html"]}},"tags":["模型更新","Hacker News 热门（buzzing.cc 中文翻译）"],"translations":{"zh-CN":{"title":"德国AI协会发布开源模型Soofi S，在英语和德语基准测试中领先","summary":"德国AI协会协调的研究联盟发布开源大语言模型Soofi S 30B-A3B。该模型总参数量316亿，每个token仅激活约32亿参数，采用Mamba-2与标准注意力层混合的MoE架构。模型完全在德国电信慕尼黑工业AI云上训练，训练数据中德语占比从第一阶段的7.2%提升至第二阶段的15.3%。在基准测试中，Soofi S在所有完全开源模型中取得英语和德语综合最高分，超越OLMo 3 32B和Apertus 70B。在HumanEval上得分73.8%，MBPP得分70.2，德语版MBPP得分84.2。上下文窗口支持最高100万token，在4万token长度下，生成吞吐量约为同规模稠密模型的8倍。模型权重已开源。","category":"模型更新","source":"The Decoder：AI News（RSS）","aggregationSource":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"德国AI协会发布开源模型Soofi S，在英语和德语基准测试中领先 - Aioga AI资讯","description":"德国AI协会协调的研究联盟发布开源大语言模型Soofi S 30B-A3B。该模型总参数量316亿，每个token仅激活约32亿参数，采用Mamba-2与标准注意力层混合的MoE架构。模型完全在德国电信慕尼黑工业AI云上训练，训练数据中德语占比从第一阶段的7.2%提升至第二阶段的15.3%。在基准测试中，Soofi S在所有完全开源模型中取得英语和德语综合最","url":"https://www.aioga.com/news/cmro1333q010ebij2087uwrac/"},"en":{"title":"The German AI Alliance releases the open-source large model Soofi S 30B-A3B, leading in benchmark tests","summary":"A research alliance coordinated by the German AI Association has released the open-source large language model Soofi S, with a total of 31.6 billion parameters, and only about 3.2 billion parameters are activated per token. This model uses a hybrid Mamba-Transformer architecture and surpasses fully open-source models such as OLMo 3 32B and Apertus 70B on German and English benchmarks.","category":"Models","source":"Hacker News 热门（buzzing.cc 中文翻译）","aggregationSource":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"The German AI Alliance releases the open-source large model Soofi S 30B-A3B, leading in benchmark tests - Aioga AI News","description":"A research alliance coordinated by the German AI Association has released the open-source large language model Soofi S, with a total of 31.6 billion parameters, and only about 3.2...","url":"https://www.aioga.com/en/news/cmro1333q010ebij2087uwrac/","contentTranslated":true,"sourceHash":"78670f5fecdc4ae3","translatedAt":"2026-07-23T01:02:45.736Z"},"ja":{"title":"ドイツAI連盟がオープンソース大規模モデルSoofi S 30B-A3Bを公開、ベンチマークテストで先行","summary":"ドイツAI協会が調整する研究連合は、オープンソースの大規模言語モデルSoofi Sを発表しました。総パラメータは316億で、1トークンあたり約32億パラメータのみがアクティブ化されます。このモデルはハイブリッドMamba-Transformerアーキテクチャを採用しており、ドイツ語と英語のベンチマークテストでOLMo 3 32BやApertus 70Bなどの完全オープンソースモデルを超えています。","category":"モデル更新","source":"Hacker News 热门（buzzing.cc 中文翻译）","aggregationSource":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"ドイツAI連盟がオープンソース大規模モデルSoofi S 30B-A3Bを公開、ベンチマークテストで先行 - Aioga AIニュース","description":"ドイツAI協会が調整する研究連合は、オープンソースの大規模言語モデルSoofi Sを発表しました。総パラメータは316億で、1トークンあたり約32億パラメータのみがアクティブ化されます。このモデルはハイブリッドMamba-Transformerアーキテクチャを採用しており、ドイツ語と英語のベンチマークテストでOLMo 3 32BやApertus 70Bなどの...","url":"https://www.aioga.com/ja/news/cmro1333q010ebij2087uwrac/","contentTranslated":true,"sourceHash":"78670f5fecdc4ae3","translatedAt":"2026-07-23T01:02:56.219Z"},"ko":{"title":"독일 AI 연합, 오픈소스 대형 모델 Soofi S 30B-A3B 출시, 벤치마크 테스트 선두","summary":"독일 AI 협회가 조정한 연구 연합이 오픈소스 대형 언어 모델 Soofi S를 발표했으며, 총 매개변수는 316억이고, 토큰당 약 32억 매개변수만 활성화됩니다. 이 모델은 혼합 Mamba-Transformer 아키텍처를 채택했으며, 독일어와 영어 벤치마크 테스트에서 OLMo 3 32B 및 Apertus 70B 등 완전 오픈소스 모델을 능가했습니다.","category":"모델 업데이트","source":"Hacker News 热门（buzzing.cc 中文翻译）","aggregationSource":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"독일 AI 연합, 오픈소스 대형 모델 Soofi S 30B-A3B 출시, 벤치마크 테스트 선두 - Aioga AI 뉴스","description":"독일 AI 협회가 조정한 연구 연합이 오픈소스 대형 언어 모델 Soofi S를 발표했으며, 총 매개변수는 316억이고, 토큰당 약 32억 매개변수만 활성화됩니다. 이 모델은 혼합 Mamba-Transformer 아키텍처를 채택했으며, 독일어와 영어 벤치마크 테스트에서 OLMo 3 32B 및 Apertus 70B 등 완전...","url":"https://www.aioga.com/ko/news/cmro1333q010ebij2087uwrac/","contentTranslated":true,"sourceHash":"78670f5fecdc4ae3","translatedAt":"2026-07-23T01:03:38.983Z"},"es":{"title":"La Alianza de IA de Alemania lanza el modelo abierto Soofi S 30B-A3B, liderando en pruebas de referencia","summary":"La alianza de investigación coordinada por la Asociación de IA de Alemania publicó el modelo de lenguaje grande de código abierto Soofi S, con un total de 31.600 millones de parámetros, activando aproximadamente 3.200 millones de parámetros por token. Este modelo utiliza una arquitectura híbrida Mamba-Transformer y supera a modelos completamente de código abierto como OLMo 3 32B y Apertus 70B en pruebas de referencia de alemán e inglés.","category":"Modelos","source":"Hacker News 热门（buzzing.cc 中文翻译）","aggregationSource":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"La Alianza de IA de Alemania lanza el modelo abierto Soofi S 30B-A3B, liderando en pruebas de referencia - Aioga Noticias de IA","description":"La alianza de investigación coordinada por la Asociación de IA de Alemania publicó el modelo de lenguaje grande de código abierto Soofi S, con un total de 31.600 millones de paráme...","url":"https://www.aioga.com/es/news/cmro1333q010ebij2087uwrac/","contentTranslated":true,"sourceHash":"78670f5fecdc4ae3","translatedAt":"2026-07-23T01:03:35.532Z"},"fr":{"title":"L'alliance allemande de l'IA publie le grand modèle open source Soofi S 30B-A3B, en tête des tests de référence","summary":"L'alliance de recherche coordonnée par l'Association allemande d'IA a publié le modèle de grande langue open source Soofi S, avec un total de 31,6 milliards de paramètres, chaque token n'activant qu'environ 3,2 milliards de paramètres. Ce modèle utilise une architecture hybride Mamba-Transformer et dépasse les modèles entièrement open source tels que OLMo 3 32B et Apertus 70B lors des tests de référence en allemand et en anglais.","category":"Modèles","source":"Hacker News 热门（buzzing.cc 中文翻译）","aggregationSource":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"L'alliance allemande de l'IA publie le grand modèle open source Soofi S 30B-A3B, en tête des tests de référence - Aioga Actualités IA","description":"L'alliance de recherche coordonnée par l'Association allemande d'IA a publié le modèle de grande langue open source Soofi S, avec un total de 31,6 milliards de paramètres, chaque t...","url":"https://www.aioga.com/fr/news/cmro1333q010ebij2087uwrac/","contentTranslated":true,"sourceHash":"78670f5fecdc4ae3","translatedAt":"2026-07-23T01:04:18.868Z"},"de":{"title":"Der deutsche KI-Verband veröffentlicht das Open-Source-Großmodell Soofi S 30B-A3B, Benchmark-Tests führend","summary":"Der vom deutschen KI-Verband koordinierte Forschungsverbund hat das Open-Source-Sprachmodell Soofi S veröffentlicht, mit insgesamt 31,6 Milliarden Parametern, wobei pro Token nur etwa 3,2 Milliarden Parameter aktiviert werden. Das Modell verwendet eine hybride Mamba-Transformer-Architektur und übertrifft in deutschen und englischen Benchmark-Tests vollständig offene Modelle wie OLMo 3 32B und Apertus 70B.","category":"模型更新","source":"Hacker News 热门（buzzing.cc 中文翻译）","aggregationSource":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"Der deutsche KI-Verband veröffentlicht das Open-Source-Großmodell Soofi S 30B-A3B, Benchmark-Tests führend - Aioga KI-News","description":"Der vom deutschen KI-Verband koordinierte Forschungsverbund hat das Open-Source-Sprachmodell Soofi S veröffentlicht, mit insgesamt 31,6 Milliarden Parametern, wobei pro Token nur e...","url":"https://www.aioga.com/de/news/cmro1333q010ebij2087uwrac/","contentTranslated":true,"sourceHash":"78670f5fecdc4ae3","translatedAt":"2026-07-23T01:04:16.977Z"},"pt-BR":{"title":"A Aliança de IA da Alemanha lançou o modelo de grande escala de código aberto Soofi S 30B-A3B, liderando em testes de benchmark","summary":"A aliança de pesquisa coordenada pela Associação de IA da Alemanha lançou o modelo de linguagem de grande escala de código aberto Soofi S, com um total de 31,6 bilhões de parâmetros, ativando apenas cerca de 3,2 bilhões de parâmetros por token. Este modelo utiliza uma arquitetura híbrida Mamba-Transformer e supera modelos totalmente abertos como OLMo 3 32B e Apertus 70B em benchmarks em alemão e inglês.","category":"模型更新","source":"Hacker News 热门（buzzing.cc 中文翻译）","aggregationSource":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"A Aliança de IA da Alemanha lançou o modelo de grande escala de código aberto Soofi S 30B-A3B, liderando em testes de benchmark - Aioga Notícias de IA","description":"A aliança de pesquisa coordenada pela Associação de IA da Alemanha lançou o modelo de linguagem de grande escala de código aberto Soofi S, com um total de 31,6 bilhões de parâmetro...","url":"https://www.aioga.com/pt-BR/news/cmro1333q010ebij2087uwrac/","contentTranslated":true,"sourceHash":"78670f5fecdc4ae3","translatedAt":"2026-07-23T01:04:56.841Z"},"ru":{"title":"Немецкий альянс ИИ выпустил открытый крупномасштабный модель Soofi S 30B-A3B, лидирующий в базовых тестах","summary":"Исследовательский альянс, координируемый Немецкой ассоциацией ИИ, выпустил открытый крупный языковой модель Soofi S с общим количеством параметров 31,6 миллиарда, при этом только около 3,2 миллиарда параметров активируются на каждый токен. Модель использует гибридную архитектуру Mamba-Transformer и превосходит полностью открытые модели, такие как OLMo 3 32B и Apertus 70B, в тестах на немецком и английском языках.","category":"模型更新","source":"Hacker News 热门（buzzing.cc 中文翻译）","aggregationSource":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"Немецкий альянс ИИ выпустил открытый крупномасштабный модель Soofi S 30B-A3B, лидирующий в базовых тестах - Aioga Новости ИИ","description":"Исследовательский альянс, координируемый Немецкой ассоциацией ИИ, выпустил открытый крупный языковой модель Soofi S с общим количеством параметров 31,6 миллиарда, при этом только о...","url":"https://www.aioga.com/ru/news/cmro1333q010ebij2087uwrac/","contentTranslated":true,"sourceHash":"78670f5fecdc4ae3","translatedAt":"2026-07-23T01:04:58.195Z"},"ar":{"title":"أصدرت رابطة الذكاء الاصطناعي الألمانية النموذج الكبير المفتوح المصدر Soofi S 30B-A3B، وتتصدر في اختبارات المعيار","summary":"أصدرت رابطة الذكاء الاصطناعي الألمانية تحالفًا بحثيًا نموذج اللغة الكبير المفتوح المصدر Soofi S، بإجمالي 31.6 مليار معلمة، حيث يتم تنشيط حوالي 3.2 مليار معلمة فقط لكل رمز. يستخدم هذا النموذج بنية هجينة من نوع Mamba-Transformer، وتفوق في اختبارات القياس الأساسية للغتين الألمانية والإنجليزية على نماذج مفتوحة المصدر بالكامل مثل OLMo 3 32B وApertus 70B.","category":"模型更新","source":"Hacker News 热门（buzzing.cc 中文翻译）","aggregationSource":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"أصدرت رابطة الذكاء الاصطناعي الألمانية النموذج الكبير المفتوح المصدر Soofi S 30B-A3B، وتتصدر في اختبارات المعيار - Aioga أخبار الذكاء الاصطناعي","description":"أصدرت رابطة الذكاء الاصطناعي الألمانية تحالفًا بحثيًا نموذج اللغة الكبير المفتوح المصدر Soofi S، بإجمالي 31.6 مليار معلمة، حيث يتم تنشيط حوالي 3.2 مليار معلمة فقط لكل رمز. يستخدم ه...","url":"https://www.aioga.com/ar/news/cmro1333q010ebij2087uwrac/","contentTranslated":true,"sourceHash":"78670f5fecdc4ae3","translatedAt":"2026-07-23T01:05:38.976Z"},"hi":{"title":"जर्मनी की एआई联盟 ने ओपन-सोर्स बड़े मॉडल Soofi S 30B-A3B जारी किया, बेंचमार्क टेस्ट में आगे","summary":"जर्मनी के AI संघ द्वारा समन्वयित शोध गठबंधन ने ओपन-सोर्स बड़े भाषा मॉडल Soofi S जारी किया, जिसमें कुल 31.6 बिलियन पैरामीटर हैं और प्रत्येक टोकन पर लगभग 3.2 बिलियन पैरामीटर ही सक्रिय होते हैं। यह मॉडल मिश्रित Mamba-Transformer आर्किटेक्चर का उपयोग करता है और जर्मन और अंग्रेजी बेंचमार्क परीक्षणों में OLMo 3 32B और Apertus 70B जैसी पूरी तरह से ओपन-सोर्स मॉडलों को पीछे छोड़ देता है।","category":"模型更新","source":"Hacker News 热门（buzzing.cc 中文翻译）","aggregationSource":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"जर्मनी की एआई联盟 ने ओपन-सोर्स बड़े मॉडल Soofi S 30B-A3B जारी किया, बेंचमार्क टेस्ट में आगे - Aioga AI समाचार","description":"जर्मनी के AI संघ द्वारा समन्वयित शोध गठबंधन ने ओपन-सोर्स बड़े भाषा मॉडल Soofi S जारी किया, जिसमें कुल 31.6 बिलियन पैरामीटर हैं और प्रत्येक टोकन पर लगभग 3.2 बिलियन पैरामीटर ही सक्रि...","url":"https://www.aioga.com/hi/news/cmro1333q010ebij2087uwrac/","contentTranslated":true,"sourceHash":"78670f5fecdc4ae3","translatedAt":"2026-07-23T01:05:41.183Z"},"it":{"title":"L'Alleanza tedesca per l'IA ha rilasciato il modello open source Soofi S 30B-A3B, al vertice nei test di riferimento","summary":"L'alleanza di ricerca coordinata dall'Associazione tedesca per l'IA ha pubblicato il modello di linguaggio di grandi dimensioni open source Soofi S, con un totale di 31,6 miliardi di parametri, attivando solo circa 3,2 miliardi di parametri per token. Questo modello utilizza un'architettura ibrida Mamba-Transformer e supera modelli completamente open source come OLMo 3 32B e Apertus 70B nei test di riferimento in tedesco e inglese.","category":"模型更新","source":"Hacker News 热门（buzzing.cc 中文翻译）","aggregationSource":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"L'Alleanza tedesca per l'IA ha rilasciato il modello open source Soofi S 30B-A3B, al vertice nei test di riferimento - Aioga Notizie IA","description":"L'alleanza di ricerca coordinata dall'Associazione tedesca per l'IA ha pubblicato il modello di linguaggio di grandi dimensioni open source Soofi S, con un totale di 31,6 miliardi...","url":"https://www.aioga.com/it/news/cmro1333q010ebij2087uwrac/","contentTranslated":true,"sourceHash":"78670f5fecdc4ae3","translatedAt":"2026-07-23T01:06:29.607Z"},"nl":{"title":"De Duitse AI-alliantie heeft het open-source grootmodel Soofi S 30B-A3B uitgebracht, met leidende prestaties bij benchmarktests","summary":"Het door de Duitse AI-vereniging gecoördineerde onderzoeksconsortium heeft het open-source grote taalmodel Soofi S uitgebracht, met in totaal 31,6 miljard parameters, waarvan slechts ongeveer 3,2 miljard parameters per token worden geactiveerd. Het model maakt gebruik van een hybride Mamba-Transformer-architectuur en overtreft volledig open-source modellen zoals OLMo 3 32B en Apertus 70B in Duits- en Engelstalige benchmarktests.","category":"模型更新","source":"Hacker News 热门（buzzing.cc 中文翻译）","aggregationSource":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"De Duitse AI-alliantie heeft het open-source grootmodel Soofi S 30B-A3B uitgebracht, met leidende prestaties bij benchmarktests - Aioga AI-nieuws","description":"Het door de Duitse AI-vereniging gecoördineerde onderzoeksconsortium heeft het open-source grote taalmodel Soofi S uitgebracht, met in totaal 31,6 miljard parameters, waarvan slech...","url":"https://www.aioga.com/nl/news/cmro1333q010ebij2087uwrac/","contentTranslated":true,"sourceHash":"78670f5fecdc4ae3","translatedAt":"2026-07-23T01:06:20.694Z"},"tr":{"title":"Almanya AI İttifakı, açık kaynaklı büyük model Soofi S 30B-A3B'yi yayımladı, kıyaslama testlerinde önde","summary":"Almanya AI Derneği tarafından koordine edilen araştırma ittifakı, açık kaynaklı büyük dil modeli Soofi S'yi yayınladı; toplam parametre sayısı 31,6 milyar ve her token başına yalnızca yaklaşık 3,2 milyar parametre aktif oluyor. Bu model, karma Mamba-Transformer mimarisi kullanıyor ve Almanca ve İngilizce kıyaslama testlerinde OLMo 3 32B ve Apertus 70B gibi tamamen açık kaynaklı modelleri geride bırakıyor.","category":"模型更新","source":"Hacker News 热门（buzzing.cc 中文翻译）","aggregationSource":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"Almanya AI İttifakı, açık kaynaklı büyük model Soofi S 30B-A3B'yi yayımladı, kıyaslama testlerinde önde - Aioga AI Haberleri","description":"Almanya AI Derneği tarafından koordine edilen araştırma ittifakı, açık kaynaklı büyük dil modeli Soofi S'yi yayınladı; toplam parametre sayısı 31,6 milyar ve her token başına yalnı...","url":"https://www.aioga.com/tr/news/cmro1333q010ebij2087uwrac/","contentTranslated":true,"sourceHash":"78670f5fecdc4ae3","translatedAt":"2026-07-23T01:07:07.805Z"},"vi":{"title":"Liên minh AI Đức phát hành mô hình lớn mã nguồn mở Soofi S 30B-A3B, đứng đầu trong các bài kiểm tra chuẩn","summary":"Liên minh nghiên cứu do Hiệp hội AI Đức phối hợp đã phát hành mô hình ngôn ngữ lớn mã nguồn mở Soofi S, với tổng số tham số là 31,6 tỷ, mỗi token chỉ kích hoạt khoảng 3,2 tỷ tham số. Mô hình này sử dụng kiến trúc lai Mamba-Transformer, vượt trội hơn các mô hình hoàn toàn mã nguồn mở như OLMo 3 32B và Apertus 70B trong các bài kiểm tra chuẩn tiếng Đức và tiếng Anh.","category":"模型更新","source":"Hacker News 热门（buzzing.cc 中文翻译）","aggregationSource":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"Liên minh AI Đức phát hành mô hình lớn mã nguồn mở Soofi S 30B-A3B, đứng đầu trong các bài kiểm tra chuẩn - Tin tức AI Aioga","description":"Liên minh nghiên cứu do Hiệp hội AI Đức phối hợp đã phát hành mô hình ngôn ngữ lớn mã nguồn mở Soofi S, với tổng số tham số là 31,6 tỷ, mỗi token chỉ kích hoạt khoảng 3,2 tỷ tham s...","url":"https://www.aioga.com/vi/news/cmro1333q010ebij2087uwrac/","contentTranslated":true,"sourceHash":"78670f5fecdc4ae3","translatedAt":"2026-07-23T01:07:07.614Z"},"id":{"title":"Aliansi AI Jerman merilis model besar sumber terbuka Soofi S 30B-A3B, memimpin dalam pengujian tolok ukur","summary":"Aliansi penelitian yang dikoordinasikan oleh Asosiasi AI Jerman merilis model bahasa besar open-source Soofi S, dengan total 31,6 miliar parameter, setiap token hanya mengaktifkan sekitar 3,2 miliar parameter. Model ini menggunakan arsitektur Mamba-Transformer hybrid, dan melampaui model open-source sepenuhnya seperti OLMo 3 32B dan Apertus 70B dalam pengujian tolok ukur bahasa Jerman dan Inggris.","category":"模型更新","source":"Hacker News 热门（buzzing.cc 中文翻译）","aggregationSource":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"Aliansi AI Jerman merilis model besar sumber terbuka Soofi S 30B-A3B, memimpin dalam pengujian tolok ukur - Berita AI Aioga","description":"Aliansi penelitian yang dikoordinasikan oleh Asosiasi AI Jerman merilis model bahasa besar open-source Soofi S, dengan total 31,6 miliar parameter, setiap token hanya mengaktifkan...","url":"https://www.aioga.com/id/news/cmro1333q010ebij2087uwrac/","contentTranslated":true,"sourceHash":"78670f5fecdc4ae3","translatedAt":"2026-07-23T01:07:51.996Z"},"th":{"title":"สมาคม AI ของเยอรมนีเปิดตัวโมเดลขนาดใหญ่แบบโอเพนซอร์ส Soofi S 30B-A3B ผลการทดสอบมาตรฐานเป็นผู้นำ","summary":"สมาคม AI ของเยอรมนีร่วมกับสหภาพวิจัยที่ประสานงานได้เผยแพร่โมเดลภาษาขนาดใหญ่โอเพ่นซอร์ส Soofi S มีพารามิเตอร์ทั้งหมด 31.6 พันล้าน พารามิเตอร์เพียงประมาณ 3.2 พันล้านที่ถูกเปิดใช้งานต่อแต่ละโทเค็น โมเดลนี้ใช้สถาปัตยกรรม Mamba-Transformer แบบผสม และสามารถทำได้เหนือกว่าโมเดลโอเพ่นซอร์สเต็มรูปแบบเช่น OLMo 3 32B และ Apertus 70B ในการทดสอบมาตรฐานเยอรมันและอังกฤษ","category":"模型更新","source":"Hacker News 热门（buzzing.cc 中文翻译）","aggregationSource":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"สมาคม AI ของเยอรมนีเปิดตัวโมเดลขนาดใหญ่แบบโอเพนซอร์ส Soofi S 30B-A3B ผลการทดสอบมาตรฐานเป็นผู้นำ - ข่าว AI Aioga","description":"สมาคม AI ของเยอรมนีร่วมกับสหภาพวิจัยที่ประสานงานได้เผยแพร่โมเดลภาษาขนาดใหญ่โอเพ่นซอร์ส Soofi S มีพารามิเตอร์ทั้งหมด 31.6 พันล้าน พารามิเตอร์เพียงประมาณ 3.2 พันล้านที่ถูกเปิดใช้งานต...","url":"https://www.aioga.com/th/news/cmro1333q010ebij2087uwrac/","contentTranslated":true,"sourceHash":"78670f5fecdc4ae3","translatedAt":"2026-07-23T01:07:52.496Z"},"pl":{"title":"Niemiecki sojusz AI opublikował otwarty model dużej skali Soofi S 30B-A3B, prowadzi w testach porównawczych","summary":"Soofi S, otwarty model językowy o dużych rozmiarach, został wydany przez konsorcjum badawcze koordynowane przez Niemieckie Stowarzyszenie AI, z całkowitą liczbą parametrów wynoszącą 31,6 miliarda, przy czym każdy token aktywuje tylko około 3,2 miliarda parametrów. Model ten wykorzystuje hybrydową architekturę Mamba-Transformer i przewyższa w testach porównawczych w języku niemieckim i angielskim w pełni otwarte modele takie jak OLMo 3 32B i Apertus 70B.","category":"模型更新","source":"Hacker News 热门（buzzing.cc 中文翻译）","aggregationSource":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"Niemiecki sojusz AI opublikował otwarty model dużej skali Soofi S 30B-A3B, prowadzi w testach porównawczych - Aioga Wiadomości AI","description":"Soofi S, otwarty model językowy o dużych rozmiarach, został wydany przez konsorcjum badawcze koordynowane przez Niemieckie Stowarzyszenie AI, z całkowitą liczbą parametrów wynosząc...","url":"https://www.aioga.com/pl/news/cmro1333q010ebij2087uwrac/","contentTranslated":true,"sourceHash":"78670f5fecdc4ae3","translatedAt":"2026-07-23T01:08:36.770Z"}}}}