[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-youdao-ziyue-4-27b-multimodal-open-source":3,"news-related-b1666218-67c1-4476-b644-34af26ea70fa":36},{"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},"b1666218-67c1-4476-b644-34af26ea70fa","有道子曰4全面开源：27B多模态模型迈入全模态时代","网易有道近日正式发布子曰大模型4.0版本，宣布核心多模态模型与TTS引擎全面开源，标志着这家教育AI老兵正式迈入全模态时代。\n\n子曰4多模态模型在27B参数规模上将视觉输入的数理能力提升至行业顶尖水平，在处理带图表的数学题、物理题等高难度视觉数理问题上表现惊艳。纯文本数理难题准确率达81.4%，同样达到行业领先。更值得关注的是，新模型采用精细化思维链重构方案，将推理思维链输出长度压缩了43.2%，这意味着可以用更少的Token、更短的推理路径给出答案，直接降低实际业务场景中的推理成本。\n\n此次一同开源的还有语音合成引擎，基于语音编码器+LLM的前沿架构，支持14种语言，3秒即可完成原声克隆，克隆准确度超过97%，相似度达85%以上。跨语种克隆不会出现口音泄露问题，这在国内TTS开源方案中相当少见。\n\n子曰团队还重构了翻译模型，引入多专家OPD模式，配合强化学习的格式奖励和语言检测机制，在提升质量的同时实现推理速度80%的提升。这对于需要高频、高并发翻译服务的产业应用场景意义重大。\n\n从最初的虚拟人口语教练Hi Echo，到如今的子曰4全模态开源，有道在教育AI领域的积累正在转化为真正的开源竞争力。对于开发者和企业而言，这套开源方案提供了一个可直接落地的高性价比选择——既能享受开源的灵活性，又有经过场景验证的性能保障。随着多模态与语音合成的门槛进一步降低，真正的生产力变革或许就在眼前。","https:\u002F\u002Fwww.infoq.cn\u002Farticle\u002Fisrd9ej6AjO6NiwAfRbI","5e4fd3d1-9cb4-44a6-bae5-9ffb449c05c1",[10,14,17,20],{"id":11,"name":12,"slug":12,"description":13,"color":13},"a8002d98-9df1-4ab9-94d4-a7625af634c4","china-ai",null,{"id":15,"name":16,"slug":16,"description":13,"color":13},"499f4b56-819d-49a3-9609-33e775143b86","multimodal",{"id":18,"name":19,"slug":19,"description":13,"color":13},"b1853a5a-d940-42b7-94f9-0488ee3f2cf7","new-model",{"id":21,"name":22,"slug":22,"description":13,"color":13},"b9bd9039-fcdb-41a8-b85b-fc1587def2b9","open-source",[24],{"id":25,"lang":26,"title":27,"summary":28,"content":13},"e084ac13-6bca-4e64-b9e3-2005f0551e44","en","Youdao Ziyue 4 open-sourced: 27B goes omni-modal","NetEase Youdao recently officially released Ziyue LLM version 4.0, announcing the full open-sourcing of its core multimodal model and TTS engine, marking this education-AI veteran as formally stepping into the full-modal era.\n\nThe Ziyue 4 multimodal model elevates the math-and-science ability of vision inputs to the industry's top level at the 27B-parameter scale, performing impressively on tough visual math\u002Fphysics problems that include charts and figures. Pure-text math problem accuracy reaches 81.4%, also industry-leading. More noteworthy, the new model adopts a refined chain-of-thought reconstruction scheme, compressing the reasoning CoT output length by 43.2% — meaning it can give answers with fewer Tokens and a shorter reasoning path, directly reducing inference cost in real business scenarios.\n\nAlso open-sourced in this release is the speech synthesis engine, based on a frontier speech-encoder + LLM architecture, supporting 14 languages, capable of completing original-voice cloning in 3 seconds with cloning accuracy above 97% and similarity above 85%. Cross-language cloning doesn't leak accent — a rare feature among domestic TTS open-source solutions.\n\nThe Ziyue team also rebuilt the translation model, introducing a multi-expert OPD mode paired with reinforcement-learning-based format reward and language detection mechanisms, improving quality while delivering 80% inference-speed boost. This is highly significant for scenarios requiring high-frequency, high-concurrency translation services.\n\nFrom the original virtual-person speaking coach Hi Echo to today's Ziyue 4 full-modal open source, Youdao's accumulated expertise in education AI is translating into real open-source competitiveness. For developers and enterprises, this open-source package offers a directly deployable, high-cost-performance choice — the flexibility of open source combined with scene-validated performance guarantees. As the bar for multimodal and speech synthesis drops further, the real productivity revolution may be just around the corner.","youdao-ziyue-4-27b-multimodal-open-source","2026-05-21T13:05:00Z","2026-05-21T13:07:43.590599Z","2026-08-19T02:08:40.142862Z",true,"agent",144,{"items":37},[38,43,48,53,58,63],{"id":39,"title":40,"news_slug":41,"published_at":42},"c677680a-a08f-420c-8107-7816827707a2","小米开源 Xiaomi-Robotics-U0：38B 具身生成统一 Tokenizer","xiaomi-robotics-u0","2026-07-14T22:10:00+00:00",{"id":44,"title":45,"news_slug":46,"published_at":47},"e3c0b314-d7b7-4901-b2b0-08ca5ef08ac7","GigaBrain-0.7开源:37k小时数据+三系统架构,世界模型进VLA决策回路","gigabrain-0-7-embodied-vla-open-source","2026-08-26T23:15:00+00:00",{"id":49,"title":50,"news_slug":51,"published_at":52},"b4214f43-353e-42e3-b48e-92dd4fc64290","京东开源 EchoWM 全模态世界模型:720p 音画同步,能跟着你走","jd-echowm-omnimodal-world-model","2026-08-25T23:10:00+00:00",{"id":54,"title":55,"news_slug":56,"published_at":57},"7ef479ae-66af-463a-802f-07a84ade93b1","商汤开源 SenseNova-U1.5-8B：原生多模态通吃生成编辑，短板全写进模型卡","sensenova-u1-5-8b-open-source-multimodal","2026-08-25T19:30:00+00:00",{"id":59,"title":60,"news_slug":61,"published_at":62},"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":64,"title":65,"news_slug":66,"published_at":67},"a151db0c-d832-4df2-ac03-2d4e58b26e99","Kimi K3 跑通 MiniTriton:Moonshot 让 LLM 第一次从零编译出自己的 GPU 编译器","kimi-k3-minitriton-gpu-compiler","2026-07-26T14:00:00+00:00"]