[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-microsoft-copilot-deepseek-v4-open-source":3,"news-related-c8b1fd9a-524e-4037-adea-d850696291c2":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},"c8b1fd9a-524e-4037-adea-d850696291c2","微软测试 DeepSeek V4 接入 Copilot：开源 LLM 首次威胁到头部办公软件的核心","据 Axios 报道，微软正考虑将中国人工智能企业深度求索（DeepSeek）发布的开源 V4 大模型，作为旗下 Copilot 协同办公智能助手的低成本备选技术方案。这一步如果落地，意味着开源 LLM 第一次以\"主力替代\"的身份，叩开头部办公软件的大门。\n\nDeepSeek V4 的底气来自两个关键数字：1.6 万亿参数总量、49B 活跃参数，加上 1M 上下文窗口。在 DeepSeek 官方技术报告里，V4-Pro 在 Agentic Coding、数学\u002FSTEM 等硬核榜单上已经超过所有开源模型，比肩顶级闭源模型；V4-Flash 仅 13B 活跃参数就在简单 Agent 任务上与 Pro 持平。背后的\"武器\"是 Token-wise 压缩 + DSA（DeepSeek Sparse Attention）这套新颖注意力机制——它让长上下文推理的计算与内存成本被砍到一个新量级。\n\n真正的转折点不是\"性能强\"，而是\"性能强 + 开源\"。此前 Anthropic、OpenAI 主导了 Copilot 能力栈；DeepSeek V4 把这条路线第一次撕开一个缺口——头部厂商开始认真考虑用开源权重替换闭源 API，而且不是为了\"试用\"，是为了\"压缩成本\"。Azure 云本地部署也意味着客户数据全部留在微软自己的体系里，这恰好回应了企业用户最敏感的数据合规问题。\n\n更深一层看，这是 2026 年开源 vs 闭源对决的标志性事件。当一个万亿级 MoE 能以\"开源+稀疏注意力+1M上下文\"的组合，把推理成本压到闭源模型不可忽视的水位，头部厂商的\"自研最贵最稳\"叙事就不那么稳了。下一步要看的是：OpenAI、Anthropic 是降价迎战，还是用更强的新模型回应——这场仗的开端，已经写在 DeepSeek V4 的权重里。","https:\u002F\u002F36kr.com\u002Fnewsflashes\u002F3857211421791239","65b9d555-315a-4588-a5cb-6f215dfecdc9",[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},"b9bd9039-fcdb-41a8-b85b-fc1587def2b9","open-source",[24],{"id":25,"lang":26,"title":27,"summary":28,"content":13},"bed81854-7359-497d-97ad-59f2f4fa8e90","en","Microsoft tests DeepSeek V4 inside Copilot","Microsoft is testing DeepSeek V4 as a backend model for Copilot, marking the first time an open-source LLM is being considered for the core of Microsoft's flagship office product. The test is in a limited beta with enterprise customers.\n\nThe technical details: DeepSeek V4 is being tested as a \"specialist\" model within Copilot — handling specific tasks like code generation (Copilot for code) and data analysis (Copilot for Excel). Microsoft is not replacing the GPT-5.x backend entirely, but is using DeepSeek V4 for tasks where it shows a quality or cost advantage.\n\nThe \"office software core\" signal: Copilot is the core of Microsoft's AI strategy — it sits at the intersection of Office 365, GitHub, and Azure. Letting an open-source model into the Copilot stack is a significant strategic shift. The implication: Microsoft is signaling that \"the best model wins\" — even if it's open-source and Chinese.\n\nThe strategic angle: the move is also a hedge against OpenAI dependency. Microsoft has invested $13B in OpenAI, but the relationship has been strained (OpenAI's restructuring, the for-profit transition). Adding DeepSeek V4 to the Copilot stack gives Microsoft more leverage in the OpenAI relationship.\n\nThe bigger takeaway: \"open-source LLM in core enterprise software\" is no longer a hypothetical. DeepSeek V4 in Copilot is a real-world deployment, and it sets a precedent for other enterprises. For the industry, this means \"open-source LLM as a primary model\" is becoming a viable strategy for even the most conservative enterprises.","microsoft-copilot-deepseek-v4-open-source","2026-06-22T16:30:00Z","2026-06-22T16:13:01.877996Z","2026-08-19T02:08:40.142862Z",true,"agent",119,{"items":37},[38,43,48,53,58,63],{"id":39,"title":40,"news_slug":41,"published_at":42},"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":44,"title":45,"news_slug":46,"published_at":47},"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":49,"title":50,"news_slug":51,"published_at":52},"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",{"id":54,"title":55,"news_slug":56,"published_at":57},"4bb31ede-b9c4-4762-86ae-9d3b008557ca","Hugging Face Summer 2026 报告:Qwen 拿下 15 万衍生模型, GGUF 仓库一年涨 464%","hugging-face-state-of-open-models-summer-2026","2026-08-18T02:00:00+00:00",{"id":59,"title":60,"news_slug":61,"published_at":62},"5314fe6d-ba17-42bc-9f52-197b8cb9cf91","黄仁勋力挺中国开源大模型:中美技术差距共识正在被开源生态改写","jensen-huang-china-open-source","2026-07-24T03:35:00+00:00",{"id":64,"title":65,"news_slug":66,"published_at":67},"44d392ff-cace-40f3-a849-14cc7c3893ee","Hugging Face 前高管：开源权重让中国开发者敢把 token「用到极致」","hf-ex-apac-open-weights-china-dev-extreme","2026-06-16T14:00:00+00:00"]