[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-tencent-hy-memory-30b-memory-writer":3,"topics-all":38,"news-related-cee60a45-f212-4a78-83dd-0226c07b3460":57},{"id":4,"title":5,"summary":6,"content":7,"original_url":8,"source_id":9,"tags":10,"translations":24,"news_slug":31,"published_at":32,"created_at":33,"modified_at":34,"is_published":35,"publish_type":36,"image_url":14,"view_count":37},"cee60a45-f212-4a78-83dd-0226c07b3460","腾讯开源记忆写入模型 Hy-Memory:30B MoE 只激活 3B,四框架平均分超 Opus 5","腾讯混元开源 Hy-Memory-30B-A3B:总参 30B、激活约 3B 的 Apache 2.0 模型,专职在 Mem0、MemOS 等框架里写长期记忆。官方评测中,它四框架六指标宏平均 55.49,超过 Claude Opus 5 的 54.62,并能迁移到未见过的 EverMemOS。","Agent 要有长期记忆，就得有模型决定\"哪些信息值得存、存成什么粒度、新证据来了怎么改旧条目\"。这份活以前大多直接塞给主力大模型——能干，但贵且大材小用。腾讯混元最近开源的 Hy-Memory-30B-A3B 换了个思路：把\"记忆写入员\"做成专才小模型，总参 30B 的 MoE，每个 token 只激活约 3B，Apache 2.0 许可，权重已挂上 Hugging Face。\n\n## 它在记忆框架里干什么\n\n外部记忆框架的写路径——Mem0、MemOS，以及腾讯自己的 Hy-Memory 框架（覆盖 Chat、Working、Coding 三类场景）——过去由通用 LLM 兼任写入员。Hy-Memory-30B-A3B 就是这块的直接替换件：决定留什么信息、用什么精度表述、新证据如何更新持久记忆库。工程侧，模型卡注明 vLLM v0.31.0 实测可部署，走 OpenAI 兼容接口。\n\n## 两阶段训练\n\n第一阶段 Memory On-Policy Distillation（记忆同策略蒸馏）：学生在真实部署的框架配置里写记忆，教师在学生写出的状态链上给稠密监督。第二阶段归因记忆强化学习：直接考察写出的记忆库，按 chunk 级回传信用，最优记忆库跨窗口继承。\n\n## 官方评测：专才小模型压过旗舰\n\n评测原则是\"只换写入模型，其余全冻结\"：框架版本、提示词、检索配置、答题模型、裁判全部固定，每个写入模型都从空记忆库开始建库；四个框架各取六项指标等权平均（LoCoMo、LongMemEval、PersonaMem-v2、Travel、Shopping、Code Memory Bench），主矩阵在固定 20% 子样本上评测。\n\n结果：Hy-Memory-30B-A3B 四框架宏平均 55.49，Claude Opus 5 为 54.62，DeepSeek V4.1 Flash 53.64，腾讯自家 Hy3 51.63，总参更大的 Qwen3.6-35B-A3B 只有 50.37。更硬的证据是训练时没见过的 EverMemOS：56.89 对 54.04，其中代码记忆基准 63.19 大幅领先对方的 49.53。\n\n## 泼点冷水\n\n这是官方自报评测，主矩阵只跑了 20% 子样本，第三方复现还没出现。逐项看也不是全面领先：Mem0 框架下的 LoCoMo，Opus 5 拿 69.47，明显高于本模型的 59.54。另外 Shopping 一栏所有写入模型都挤在约 14-24 分的低区间，偏好类记忆对谁都是硬骨头。\n\n## 所以呢\n\n这次发布真正值得记的不是\"3B 赢了旗舰\"，而是它验证了一个工程判断：记忆写入是窄而深的任务，专才小模型比通用大模型更划算——旗舰 API 当记忆写入员，成本和延迟都是奢侈品。技术报告在 arXiv:2610.22196，代码开在 GitHub 的 Tencent-Hunyuan\u002FHy-Memory。对做 Agent 记忆系统的团队，[模型卡](https:\u002F\u002Fhuggingface.co\u002Ftencent\u002FHy-Memory-30B-A3B)就是低成本换件实验的入场券。","https:\u002F\u002Fhuggingface.co\u002Ftencent\u002FHy-Memory-30B-A3B","19812099-38b5-4f44-8bf6-54f72dbb1144",[11,15,18,21],{"id":12,"name":13,"slug":13,"description":14,"color":14},"01598627-1ea6-4b27-a5d8-874971571a71","llm",null,{"id":16,"name":17,"slug":17,"description":14,"color":14},"7e89b5cc-57db-4f37-bc6d-28919a73931c","model-release",{"id":19,"name":20,"slug":20,"description":14,"color":14},"d11f0044-8aef-487c-bebe-89ce4683a4a3","moe",{"id":22,"name":23,"slug":23,"description":14,"color":14},"b9bd9039-fcdb-41a8-b85b-fc1587def2b9","open-source",[25],{"id":26,"lang":27,"title":28,"summary":29,"content":30},"fd68a92e-67e7-4651-b0a3-c5e292762cc9","en","Tencent open-sources Hy-Memory, a 3B-active memory-writer MoE","Tencent open-sources Hy-Memory-30B-A3B: an Apache-2.0 MoE, 3B active, that writes memory for Mem0\u002FMemOS frameworks and beats Claude Opus 5 on average.","Long-term memory for agents needs a model to decide what information is worth keeping, at what granularity, and how new evidence should update old entries. That write path has usually been handed to the main general-purpose LLM — capable, but expensive and overqualified. Tencent Hunyuan's newly open-sourced Hy-Memory-30B-A3B takes a different route: it turns the \"memory writer\" into a specialist, a 30B-total-parameter MoE that activates only about 3B per token, under an Apache 2.0 license with weights on Hugging Face.\n\n## What it does inside memory frameworks\n\nThe write path of external memory frameworks — Mem0, MemOS, and Tencent's own Hy-Memory framework covering Chat, Working, and Coding scenarios — used to be staffed by a general-purpose LLM. Hy-Memory-30B-A3B is a direct replacement for that slot: it decides what to retain, how precisely to represent it, and how new evidence updates the persistent memory bank. On the engineering side, the model card notes deployment is tested with vLLM v0.31.0 behind an OpenAI-compatible API.\n\n## Two-stage training\n\nStage one is Memory On-Policy Distillation: the student writes memory inside the same deployed harness configurations, while the teacher supplies dense supervision on the student-written state chains. Stage two is Attributed Memory Reinforcement Learning: the resulting memory banks are examined directly, credit is assigned at the chunk level, and the best bank is carried across windows.\n\n## Official eval: the small specialist edges out the flagship\n\nThe evaluation changes only the memory writer — harness version, prompts, retrieval configuration, answer model, and judge are all held fixed, and every writer builds its bank from an empty state. Each of the four harnesses (Mem0, Hy-Memory, MemOS, EverMemOS) is scored on an equal-weight average of six metrics — LoCoMo, LongMemEval, PersonaMem-v2, Travel, Shopping, and Code Memory Bench — with the main matrix evaluated on a fixed 20% subsample.\n\nHy-Memory-30B-A3B posts a four-harness macro average of 55.49, ahead of Claude Opus 5 at 54.62, DeepSeek V4.1 Flash at 53.64, Tencent's own Hy3 at 51.63, and the larger Qwen3.6-35B-A3B at 50.37. The stronger evidence is EverMemOS, held out from training: 56.89 versus 54.04, including a wide 63.19-to-49.53 lead on the code-memory benchmark.\n\n## Caveats\n\nThis is a vendor-reported evaluation on 20% subsamples, with no third-party replication yet. Nor is it a blanket win: on LoCoMo under the Mem0 harness, Opus 5 scores 69.47 against this model's 59.54. And every writer sits in a low band of roughly 14-24 on the Shopping metric, suggesting preference-style memory is hard across the board.\n\n## So what\n\nThe takeaway is not \"3B beats a flagship\" clickbait; it is a validated engineering judgment — memory writing is a narrow, deep task where a small specialist beats a general-purpose giant on cost and latency alike. The tech report is at arXiv:2610.22196 and the code lives in Tencent-Hunyuan\u002FHy-Memory on GitHub. For teams building agent memory systems, the [model card](https:\u002F\u002Fhuggingface.co\u002Ftencent\u002FHy-Memory-30B-A3B) is a ticket to a cheap swap-in experiment.","tencent-hy-memory-30b-memory-writer","2026-10-17T15:14:21Z","2026-10-17T15:14:54.497743Z","2026-10-17T15:14:54.497754Z",true,"agent",221,[39,48],{"slug":40,"tag_slug":40,"title_zh":41,"title_en":42,"intro_zh":43,"intro_en":44,"id":45,"is_active":35,"created_at":46,"modified_at":47},"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":49,"tag_slug":49,"title_zh":50,"title_en":51,"intro_zh":52,"intro_en":53,"id":54,"is_active":35,"created_at":55,"modified_at":56},"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":58},[59,64,69,74,79,84],{"id":60,"title":61,"news_slug":62,"published_at":63},"b44fb160-9027-44cd-b355-a66869004abc","Nemotron 3.5 Super VL 开源:120B 总参只激活 12B","nemotron-3-5-super-vl-open-120b-moe","2026-10-20T17:11:56+00:00",{"id":65,"title":66,"news_slug":67,"published_at":68},"bbfcbee4-8d92-4078-bb1e-eaf4d1c1e146","Beam 权重落地:501B MoE 三精度齐发","beam-1-0-weights-open-agent-stack","2026-10-18T21:10:09+00:00",{"id":70,"title":71,"news_slug":72,"published_at":73},"d529114c-bfd7-4efb-9af4-444d3065b397","Step 5 Preview 开源,2.7 美元\u002F百万输出","stepfun-step-5-preview-600b-moe","2026-10-17T21:12:04+00:00",{"id":75,"title":76,"news_slug":77,"published_at":78},"ac666c4e-4756-43d8-840a-4eacbf88f804","Meta 开源 816B MoE 模型 Muse Spark Open,跑在两台 Mac Studio 上","meta-muse-spark-open-816b-mac-studio","2026-10-17T03:00:00+00:00",{"id":80,"title":81,"news_slug":82,"published_at":83},"1b6ec126-cba1-47da-870b-da19cc523763","Meta 开源 816B MoE 模型 Muse Spark Open:1M 上下文,两台 Mac Studio 跑得动","meta-muse-spark-open-816b-moe-1m-mac-studio","2026-10-15T11:00:00+00:00",{"id":85,"title":86,"news_slug":87,"published_at":88},"f3c43720-650b-4054-9349-a1386e06c8fc","Le Chonk 把法国拉回非美\u002F美头部:38 分的 Mistral Large 4","mistral-large-4-le-chonk-intelligence-index-38","2026-10-08T03:30:00+00:00"]