[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-mach-mind-4-flash-35b-moe":3,"topics-all":36,"news-related-75318561-aa25-43fe-965f-37e61d2cf295":65},{"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},"75318561-aa25-43fe-965f-37e61d2cf295","理想汽车 Mach-Mind-4-Flash 技术报告：35B MoE 用 3B 激活把千亿级模型拉下马","理想汽车基础模型团队发布 Mach-Mind-4-Flash 技术报告——只靠后训练阶段的精修，就让 35B 总参数 \u002F 3B 激活参数的 MoE 模型在多项基准上追平甚至反超千亿参数级对手。流水线分三段：第一阶段搭建统一的 RL\u002FOPD 训练基础设施，靠动态多教师调度和算子级加速把端到端训练速度推高 17%；第二阶段用 Multi-Teacher On-Policy Distillation（MOPD）把 Reasoning\u002FGeneral\u002FAgent 三路领域 RL 专家并行训练，再用 routed reverse-KL 目标融合成单一通才，绕开\"混合奖励跷跷板\"的退化陷阱；第三阶段是 Hybrid Median-length Policy Optimization（HMPO），单阶段内把推理链长度压缩 19–46%，精度损失 ≤0.7pp。最终跑出 AIME'26 92.70、IFBench 82.82、Behavioral-SafetyBench 80.74、BFCL-v4 75.80、BrowseComp-zh 72.31、ClawBench 84.20 等成绩——在智能体、推理、安全多个维度上以 10–30 倍更小的激活参数打败同级对手，把\"参数堆叠\"叙事进一步压缩。","https:\u002F\u002Farxiv.org\u002Fabs\u002F2607.09375","7437aeb9-930c-4866-a2e9-48003c1a792b",[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},"01598627-1ea6-4b27-a5d8-874971571a71","llm",{"id":18,"name":19,"slug":19,"description":13,"color":13},"7e89b5cc-57db-4f37-bc6d-28919a73931c","model-release",{"id":21,"name":22,"slug":22,"description":13,"color":13},"b1853a5a-d940-42b7-94f9-0488ee3f2cf7","new-model",[24],{"id":25,"lang":26,"title":27,"summary":28,"content":13},"577d2f93-5a47-449c-9bbf-7f4adc212834","en","Li Auto's Mach-Mind-4-Flash: 35B MoE punches above its weight","Li Auto's Foundation Model team released the Mach-Mind-4-Flash tech report — relying only on post-training refinement, a 35B-total \u002F 3B-activated MoE model catches up to or even surpasses trillion-parameter-tier opponents across multiple benchmarks. The pipeline has three stages: in stage one, a unified RL\u002FOPD training infrastructure is built, and dynamic multi-teacher scheduling plus operator-level acceleration push end-to-end training speed up 17%; in stage two, Multi-Teacher On-Policy Distillation (MOPD) is used to train three domain RL experts (Reasoning \u002F General \u002F Agent) in parallel, then fuse them into a single generalist with a routed reverse-KL objective, side-stepping the \"mixed-reward seesaw\" degradation trap; stage three is Hybrid Median-length Policy Optimization (HMPO), which compresses reasoning chain length by 19–46% within a single stage, with accuracy loss ≤0.7pp. In the end it runs out AIME'26 92.70, IFBench 82.82, Behavioral-SafetyBench 80.74, BFCL-v4 75.80, BrowseComp-zh 72.31, ClawBench 84.20 — beating same-tier opponents on agent, reasoning, and safety dimensions with 10–30× smaller activated parameters, further compressing the \"parameter stacking\" narrative.","mach-mind-4-flash-35b-moe","2026-07-13T10:08:00Z","2026-07-13T02:11:17.616673Z","2026-08-19T02:08:40.142862Z",true,"agent",380,[37,47,56],{"slug":38,"tag_slug":39,"title_zh":40,"title_en":41,"intro_zh":42,"intro_en":43,"id":44,"is_active":33,"created_at":45,"modified_at":46},"image-generation","text-to-image","图像生成：从文生图到逐框编辑的模型竞赛","Image Generation: from text-to-image to frame-level editing","文生图已是红海，战场正转向编辑与可控性：FLUX.2\u002F3 用开源和量化把生图搬进消费级显卡，Qwen-Image、Seedream 等国产模型紧咬不放，GPT Image 2 则把推理能力带进像素。本专题追踪图像生成与编辑模型的发布、架构演进、加速量化与商业化数据，看「好图」如何变成「可用的图」。","Text-to-image is a crowded race, and the frontier has moved to editing and control: FLUX.2\u002F3 push open weights and quantization onto consumer GPUs, Qwen-Image and Seedream close in from China, and GPT Image 2 brings reasoning into pixels. This topic tracks model releases, architecture shifts, acceleration work, and the business numbers behind \"pretty\" becoming \"usable\".","52de7c59-7864-4569-8ae7-3d656d26a1a1","2026-11-20T07:20:23.534506Z","2026-11-20T07:20:23.534527Z",{"slug":48,"tag_slug":48,"title_zh":49,"title_en":50,"intro_zh":51,"intro_en":52,"id":53,"is_active":33,"created_at":54,"modified_at":55},"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":57,"tag_slug":57,"title_zh":58,"title_en":59,"intro_zh":60,"intro_en":61,"id":62,"is_active":33,"created_at":63,"modified_at":64},"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":66},[67,72,77,82,87,92],{"id":68,"title":69,"news_slug":70,"published_at":71},"7df8de00-66e7-4969-896b-c8d2b6da3e08","阿里发布 Qwen 3.7 Max：首个闭源旗舰模型冲击前沿，百万 token 上下文刷新纪录","qwen-3-7-max-closed-source-flagship-1m-35h","2026-05-30T10:01:00+00:00",{"id":73,"title":74,"news_slug":75,"published_at":76},"caa54bff-d57a-411c-9dae-43f1d4d46875","DeepSeek V4 重磅登场：长期记忆技术突破重塑AI能力边界","deepseek-v4-engram-ltm-long-term-memory-87pct","2026-04-22T07:05:00+00:00",{"id":78,"title":79,"news_slug":80,"published_at":81},"f565e355-1269-4152-91ad-3c3810659d27","月之暗面主动砍掉 9.5 个点:开放权重先给自己设限","kimi-k3-1-pro-cyber-capability-cap","2026-10-27T07:30:00+00:00",{"id":83,"title":84,"news_slug":85,"published_at":86},"3c43418d-60d9-4bdc-bfaa-a525610ead00","月之暗面 K3.1 Pro:不换基座,后训练涨 11 分","kimi-k3-1-pro-launch","2026-10-23T23:08:50+00:00",{"id":88,"title":89,"news_slug":90,"published_at":91},"e304fe91-16d5-4c28-81c7-0c4c0dd8c669","小米公开MiMo-V2.6训练账本:RL烧了260万美元","xiaomi-mimo-v2-6-rl-scaling-ledger","2026-10-10T13:11:49+00:00",{"id":93,"title":94,"news_slug":95,"published_at":96},"02d83474-ba8d-402e-b4e3-be117be6a91d","Eleven v4 发布:新架构登顶 TTS 盲测榜,推理延迟 100ms","eleven-v4-tts-speech-arena","2026-09-29T13:07:54+00:00"]