[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-tencent-hunyuan-foundation-model-dept":3,"topics-all":36,"news-related-bd050dd6-4d85-4616-a004-55c23c533a24":55},{"id":4,"title":5,"summary":6,"content":6,"original_url":7,"source_id":8,"tags":9,"translations":23,"news_slug":29,"published_at":30,"created_at":31,"modified_at":32,"is_published":33,"publish_type":34,"image_url":13,"view_count":35},"bd050dd6-4d85-4616-a004-55c23c533a24","腾讯混元合并大语言模型与多模态团队，成立基础模型部探索全模态统一","腾讯宣布将混元大语言模型部门与多模态模型部门合并,组建统一的基础模型部,由首席AI科学家姚顺雨负责,目标是打通文本、图像、视频、音频之间的能力壁垒,探索\"全模态模型\"的智能上限。\n\n这一组织调整背后是国内头部厂商对AGI路径选择的集体押注。当下业界共识是:单一模态的预训练范式已触顶,下一步必须靠跨模态的联合训练和统一表征才能继续突破。但\"全模态模型\"也是一条比单纯做LLM更难的路——不同模态的token分布、训练节奏、评测基准差异巨大,统一架构很容易变成\"什么都沾一点但什么都做不精\"。\n\n值得注意的是,这并非腾讯第一次做混元团队调整。今年7月就有报道指出混元撤出多模态理解、把资源押给世界模型;此次合并又被解读为\"重新拥抱多模态\"。但细看合并的方向——把LLM团队也并入——意味着腾讯想做的不是\"LLM+多模态插件\",而是把多模态当作基础能力融入预训练核心。\n\n对国内大模型竞争格局而言,这是一次清晰的战略表态:在DeepSeek V4和GLM系列持续占据开源榜单前列、阿里通义走\"模型即API\"路线的当下,腾讯选了最重的一条路——用组织架构的对齐来换取技术路线的统一。短期看,合并会带来阵痛,两个团队的工程管线、评测体系、数据资产都需要重新磨合;但中长期,如果全模态确实是AGI的必经之路,这种\"提前一体化\"的组织成本反而可能成为腾讯的护城河。\n\n说到底,做基础模型从来不是一场冲刺,而是一场马拉松。组织架构先对齐,技术路线才有可能跑通。","https:\u002F\u002F36kr.com\u002Fnewsflashes\u002F3909248625513863","d46ec0a7-501b-4ef8-9c89-2391b2701b3b",[10,14,17,20],{"id":11,"name":12,"slug":12,"description":13,"color":13},"5e628969-6d2a-437f-998a-104e4b16cfb1","ai-progress",null,{"id":15,"name":16,"slug":16,"description":13,"color":13},"a8002d98-9df1-4ab9-94d4-a7625af634c4","china-ai",{"id":18,"name":19,"slug":19,"description":13,"color":13},"01598627-1ea6-4b27-a5d8-874971571a71","llm",{"id":21,"name":22,"slug":22,"description":13,"color":13},"499f4b56-819d-49a3-9609-33e775143b86","multimodal",[24],{"id":25,"lang":26,"title":27,"summary":28,"content":28},"4009d98c-6445-46f6-b3e5-1f07df50226b","en","Tencent Hunyuan merges LLM and multimodal teams","Tencent announced it is merging its Hunyuan large language model department and its multimodal model department into a single Foundation Model department, led by chief AI scientist Yao Shunyu, with the goal of breaking the capability walls between text, image, video, and audio and exploring the upper limit of intelligence in a \"unified all-modality model\". Behind this organizational shift is a collective bet by top Chinese vendors on the AGI path. The current industry consensus: single-modality pretraining paradigms have hit their ceiling, and the next step must come from cross-modality joint training and unified representations. But \"unified all-modality\" is also a harder path than simply doing LLMs — different modalities differ hugely in token distribution, training cadence, and evaluation benchmarks, and a unified architecture can easily become \"a bit of everything but master of none\". Worth noting: this isn't Tencent's first Hunyuan team reshuffle. In July there was already a report that Hunyuan exited multimodal understanding and bet the bullet on world models; this merger is being read as \"re-embracing multimodal\". But look closer at the direction of the merger — bringing the LLM team in too — and what Tencent wants is not \"LLM + multimodal plugins\", but to fold multimodality into the pretraining core as a foundational capability. For China's large-model competitive landscape, this is a clear strategic statement: with DeepSeek V4 and the GLM series consistently leading the open-source leaderboards, and Alibaba Tongyi going the \"model as API\" route, Tencent has chosen the heaviest path — using organizational alignment to buy technical-route unification. In the short term, the merger will bring pain — both teams' engineering pipelines, evaluation systems, and data assets all need to be re-aligned. But in the medium-to-long term, if all-modality really is the necessary path to AGI, this \"early integration\" organizational cost could end up being Tencent's moat. At the end of the day, building foundation models has never been a sprint — it's a marathon. Organizational alignment comes first; only then can the technical route have a chance of working.","tencent-hunyuan-foundation-model-dept","2026-07-24T03:00:00Z","2026-07-24T14:02:53.271297Z","2026-08-19T02:08:40.142862Z",true,"agent",173,[37,46],{"slug":38,"tag_slug":38,"title_zh":39,"title_en":40,"intro_zh":41,"intro_en":42,"id":43,"is_active":33,"created_at":44,"modified_at":45},"ai-for-science","AI for Science 2026：从 UniPert 到 GPT-Rosalind 的硬核进化","AI for Science 2026: from UniPert to GPT-Rosalind","生命科学、化学材料、物理世界模型——AI 正在从\"语言工具\"变成\"实验伙伴\"。本专题收录 AI 在三大科学方向的关键节点：UniPert 统一基因与化学扰动空间、GPT-Rosalind 端到端生命科学推理、达摩院 AI 智能体 28 小时找到 4 种超导新材料、Anthropic Claude Science 把工作台做成标准品。","From language tool to lab partner — AI is reshaping life sciences, chemistry\u002Fmaterials, and physical world models. This topic covers the key milestones: UniPert unifying genetic-chemical perturbation spaces, GPT-Rosalind's end-to-end life-sciences reasoning, DAMO's AI agent discovering 4 superconducting materials in 28 hours, and Anthropic's Claude Science workbench going mainstream.","988a4300-5fab-41c4-b5d8-63711a2dc757","2026-09-10T01:34:15.296649Z","2026-09-10T01:34:15.296663Z",{"slug":47,"tag_slug":47,"title_zh":48,"title_en":49,"intro_zh":50,"intro_en":51,"id":52,"is_active":33,"created_at":53,"modified_at":54},"h3-series","MiniMax H3 系列：从开源权重到 35 倍吞吐","MiniMax H3 Series: from open weights to 35x throughput","MiniMax H3 自 2026 年 8 月开源以来节奏密集：官方把生成、参考与编辑收回一个模型；ComfyUI 当天压进 RTX 3060；摩尔线程 3 小时完成国产 GPU 适配；fal 后训练版把吞吐拉到 35 倍；FastH3 蒸馏再砍推理成本。本专题持续追踪 H3 的发布—开源—蒸馏—部署全链路。","Since MiniMax open-sourced H3 in August 2026 the pace has been relentless: one unified omni-modal model, same-day ComfyUI support down to an RTX 3060, a 3-hour Day-0 port to Moore Threads GPUs, fal's post-trained H3 Max at 35x throughput, and FastH3 distillation cutting inference cost further. This topic tracks the full H3 chain — release, open weights, distillation, deployment.","83ef0daa-3c31-4cb3-86ed-e5ee58654d5f","2026-09-08T07:33:19.942193Z","2026-09-08T07:33:19.942209Z",{"items":56},[57,62,67,72,77,82],{"id":58,"title":59,"news_slug":60,"published_at":61},"63506aeb-566f-47dc-a59f-d4f0197e4545","面壁 MiniCPM 拿下三星旗舰端:7 款端侧 LLM 同日备案,标志端侧 AI 进入合规量产节奏","minicpm-samsung-end-side","2026-07-15T14:01:00+00:00",{"id":63,"title":64,"news_slug":65,"published_at":66},"6b349d2c-3d03-4cb9-8f47-63e8c288d0db","美方三机构联合指控六家中国 AI 企业系统性蒸馏美国模型","us-accuses-six-chinese-ai-firms-of-distillation","2026-09-10T01:08:40+00:00",{"id":68,"title":69,"news_slug":70,"published_at":71},"0d8fdf45-4585-47c0-9e78-3652e318b156","Apple Intelligence 中国版落地:通义千问接管语言 AI,百度负责视觉搜索","apple-intelligence-china-qwen-baidu-2026","2026-08-25T12:00:00+00:00",{"id":73,"title":74,"news_slug":75,"published_at":76},"1844afb1-3a1c-4acd-9e4c-f5e2792a2018","下载免费不等于商用免费：HF Summer 2026 隐藏的开源前沿许可证分水岭","frontier-license-shift-hf-summer-2026","2026-08-23T12:30:00+00:00",{"id":78,"title":79,"news_slug":80,"published_at":81},"2fc64783-8b2a-49a3-939b-edf02bff3622","Ox Alpha 指纹指向 GLM-5.3:OpenRouter 的 1M 上下文隐身模型可能是智谱","ox-alpha-glm-5-3-stealth-zhipu","2026-08-22T14:00:00+00:00",{"id":83,"title":84,"news_slug":85,"published_at":86},"9389d1ed-dd2d-41cb-bbc5-9a543e2b2f71","开源报告里的「参数天花板」分水岭:中国实验室把上限拉到2.78T,美国还在130B徘徊","hf-summer-2026-china-open-weight-parameter-ceiling","2026-08-20T06:00:00+00:00"]