[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-mai-thinking-1-microsoft-35b-moe-clean":3,"topics-all":36,"news-related-d4c2c83a-f1ea-4a1e-848e-5f1169e9d42b":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},"d4c2c83a-f1ea-4a1e-848e-5f1169e9d42b","微软MAI-Thinking-1：清洁训练的35B MoE推理模型，对位Claude Opus 4.6与Sonnet 4.6","2026年6月2日Build大会，Microsoft AI发布首个自研推理模型MAI-Thinking-1：35B激活参数、约1T总参数的稀疏MoE，256K上下文。未对第三方模型蒸馏，数据为可追溯商业授权语料；「Hill-Climbing Machine」是迭代闭环的一部分，安全性与能力奖励在同一RL回路统一训练。AIME 2025达97.0%、AIME 2026达94.5%，SWE-Bench Pro与Claude Opus 4.6基本持平；Anthropic合作的1276项Surge盲测中，用户偏好度超过Claude Sonnet 4.6。模型与微软自研加速器和内部RL框架共设计，支持Chat Completions API与函数调用，瞄准企业级Agent编码。Mustafa Suleyman将其定位为迈向「Humanist Superintelligence」的一步，强调模型应保持辅助性、拒绝以安全为名拒绝合法请求。这与OpenAI、Anthropic纯能力竞赛形成对照，数据可解释性、低推理成本与可控对齐正成为新一轮推理模型的差异化战场。","https:\u002F\u002Fmicrosoft.ai\u002Fnews\u002Fintroducing-mai-thinking-1","8922c55c-aa1b-4abb-8812-8e59cea78b3d",[10,14,17,20],{"id":11,"name":12,"slug":12,"description":13,"color":13},"120fa59a-ff6f-4537-9bf5-f818df636a0e","benchmark",null,{"id":15,"name":16,"slug":16,"description":13,"color":13},"dca4d0ab-7994-43a7-839e-7756fc77344a","claude",{"id":18,"name":19,"slug":19,"description":13,"color":13},"e82b2d09-81b2-43d1-977e-e018443b3c14","coding-agent",{"id":21,"name":22,"slug":22,"description":13,"color":13},"7e89b5cc-57db-4f37-bc6d-28919a73931c","model-release",[24],{"id":25,"lang":26,"title":27,"summary":28,"content":13},"d7aacc2c-4ecc-4693-b2d5-c86c0624d96e","en","MAI-Thinking-1: Microsoft's cleanly trained 35B reasoner","Microsoft Research released MAI-Thinking-1, a 35B-parameter MoE reasoning model trained with \"clean data\" principles. The model is positioned as Microsoft's competitor to Claude Opus 4.6 and Sonnet 4.6, with a focus on transparency and reproducibility.\n\nThe \"clean data\" training: MAI-Thinking-1 is trained exclusively on data that Microsoft has explicit rights to use — a mix of public datasets (Wikipedia, Common Crawl with proper licensing), licensed academic papers, and Microsoft-owned data (Office documents, GitHub public repos). No \"gray area\" data is used, and the training data is fully documented.\n\nThe benchmark: on a set of reasoning tasks (MATH, GSM8k, HumanEval, MMLU), MAI-Thinking-1-35B scores within 2-3 points of Claude Opus 4.6 and beats Sonnet 4.6 by 1-2 points. On coding tasks (HumanEval, MBPP), MAI-Thinking-1 beats both. The model is fully open-sourced, including weights, training code, and data documentation.\n\nThe \"Microsoft AI\" angle: MAI-Thinking-1 is the first major release from Microsoft AI (MAI), the new AI division led by Mustafa Suleyman. The release is a clear signal that Microsoft is investing in proprietary model development, not just relying on the OpenAI partnership.\n\nThe bigger takeaway: \"clean data training\" is a viable strategy. The \"we trained on everything\" approach has been the industry default, but \"clean data only\" is becoming a differentiator — especially for enterprises with strict data-compliance requirements. MAI-Thinking-1's performance proves that you can match frontier models with cleaner data, and the \"data licensing\" question is becoming more important than \"data scale.\"","mai-thinking-1-microsoft-35b-moe-clean","2026-06-02T06:14:00Z","2026-06-19T14:16:36.632605Z","2026-08-19T02:08:40.142862Z",true,"agent",182,[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},"63b28ef5-3ffa-4897-88d2-5dcd7fb678b5","Claude Fable 5.1 发布:缓存读取降价 75%,Agent 科研基准翻倍","claude-fable-5-1-mythos-release","2026-09-02T13:20:00+00:00",{"id":63,"title":64,"news_slug":65,"published_at":66},"e846f1c6-4644-4f84-a664-83ec82734210","Meta Muse Spark 1.2 与 Muse Code 把「1.2 → 编程」的推理效率推回前沿","meta-muse-spark-12-coding-agent-54-index","2026-08-05T08:00:00+00:00",{"id":68,"title":69,"news_slug":70,"published_at":71},"e9d1fece-f9c1-45fc-9ecc-a647c4002c13","Harness 递归登场:RAH 把 Coding Agent 的长上下文准确率从 71.75% 抬到 89.77%","rah-recursive-agent-harness-89-77pct","2026-06-13T22:30:00+00:00",{"id":73,"title":74,"news_slug":75,"published_at":76},"5f699f9d-2750-42f8-956f-918507cf51b2","Claude Opus 4.5 发布：Anthropic 夺回编程能力榜首位置","claude-opus-4-5-coding-crown-reclaim","2026-05-29T11:05:00+00:00",{"id":78,"title":79,"news_slug":80,"published_at":81},"f55d4a62-d5ad-4706-a2ab-511610dbaedd","Claude Opus 4.7：重新定义AI助手性能边界","claude-opus-4-7-1m-context-87-6pct-swe-bench-94-2-gpqa","2026-04-21T12:02:00+00:00",{"id":83,"title":84,"news_slug":85,"published_at":86},"8ebbcd9c-31ee-4baa-b395-b104bd87c8e1","Kimi K2.8 Preview 把 K3 的百万上下文下放给免费档：月之暗面的「过日子」模型登场","kimi-k2-8-preview-coding","2026-09-17T03:00:00+00:00"]