[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-minicpm5-1b-openbmb-17-9-intelligence":3,"topics-all":36,"news-related-d6ee0dbd-e2b3-4558-9b11-75f1b2b8fb25":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},"d6ee0dbd-e2b3-4558-9b11-75f1b2b8fb25","MiniCPM5-1B：1B级新王登基，OpenBMB用17.9智能指数重写小模型效率边界","OpenBMB 近日发布 MiniCPM5-1B，作为 MiniCPM5 系列首个 checkpoint，它以 10 亿参数的紧凑规模在 Artificial Analysis Intelligence Index 上拿下 17.9 分，刷新 1B 及以下开源模型纪录。比次优的 Qwen3.5 0.8B 推理版（10.5）高出 7.4 分，也以不到一半的参数压制了 Qwen3.5 2B（16.3）。相比上一代 MiniCPM-V 4.6 1.3B（12.7），新模型参数减少约 23% 的同时智能反涨 5 分，把小模型 Pareto 前沿又往左上推了一格。\n\n训练范式上，后训练被拆成 SFT、On-Policy Distillation（OPD）和 RL 三段：先做有监督微调，再让模型在自采样轨迹上蒸馏，最后用强化学习打磨。OPD 让模型在训练时就能\"自己思考\"，无需冷启动蒸馏数据；RL 阶段在 OPD 把分布校准到接近 SFT 模型后，规避了常见塌缩。这是这次 1B 模型能跑出 17.9 智能分的关键。\n\n能力上，模型内置 `\u003Cthink>` 模板和 `enable_thinking` 开关，让同一 checkpoint 既能当\"秒回助理\"也能当\"深思熟虑的推理者\"。代码与智能体能力尤其突出，42.57 的综合均分明显领先同体量对手。AA-Omniscience 拿到 -1 分，是同尺寸最高——它宁可拒绝回答也不瞎猜，比硬卷答题率更贴近真实部署需求。\n\n部署侧，官方同步放出 BF16、GGUF、MLX 三种格式，128K 上下文窗口 + 1B 参数的组合让笔记本、手机、桌面宠物都跑得动。\n\n把视野拉远看，1B 级逆袭意味着三件事：端侧 Agent 真正具备\"类主机\"推理能力；AI 部署成本曲线继续下探；\"参数越大越好\"的旧叙事正被\"训练后+蒸馏+RL\"组合拳瓦解。MiniCPM5-1B 证明只要训练后流程做扎实，1B 也能承担生产级 Agent 任务，本地 1B 可能不再是\"凑合用\"的备胎，而是低延迟、高隐私场景的第一选择。","https:\u002F\u002Fgithub.com\u002Fopenbmb\u002Fminicpm","71df6775-935e-4b09-bda9-e03ee3eb8191",[10,14,17,20],{"id":11,"name":12,"slug":12,"description":13,"color":13},"7ac06d8e-b074-4147-abfc-ffaa4c6b8744","ai-efficiency",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},"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},"0d27b971-0f73-4095-9a61-4df57b636822","en","MiniCPM5-1B: the new 1B king, 17.9 on the intelligence index","OpenBMB recently released MiniCPM5-1B, the first checkpoint in the MiniCPM5 series. With a compact 1B parameter scale, it scored 17.9 on the Artificial Analysis Intelligence Index, refreshing the open-source record at 1B and below. It is 7.4 points higher than the next-best Qwen3.5 0.8B inference version (10.5), and crushes Qwen3.5 2B (16.3) with less than half the parameters. Compared to the previous generation MiniCPM-V 4.6 1.3B (12.7), the new model reduces parameters by about 23% while intelligence rises by 5 points, pushing the small-model Pareto frontier another notch to the upper left.\n\nOn the training paradigm side, post-training is split into three stages: SFT, On-Policy Distillation (OPD), and RL — supervised fine-tuning first, then letting the model distill on self-sampled trajectories, and finally polishing with reinforcement learning. OPD lets the model \"think for itself\" during training without the need for cold-start distillation data; in the RL stage, after OPD calibrates the distribution close to the SFT model, common collapse is avoided. This is the key to the 1B model achieving 17.9 intelligence.\n\nOn the capability side, the model has a built-in `\u003Cthink>` template and `enable_thinking` switch, letting the same checkpoint act both as a \"fast-response assistant\" and as a \"deliberate reasoner.\" Code and agent capability are particularly outstanding, with a comprehensive average of 42.57 clearly leading peers of the same size. AA-Omniscience gets -1, the highest at the same size — it would rather refuse to answer than guess, which is closer to real deployment needs than hard-scoring answer rates.\n\nOn the deployment side, the official release simultaneously provides BF16, GGUF, and MLX three formats, and the combination of 128K context window + 1B parameters lets laptops, phones, and desktop pets all run.\n\nPulling the view back, the 1B-class upset means three things: on-device Agents truly gain \"class-mainframe\" reasoning capability; AI deployment cost curves continue to fall; and the old narrative of \"bigger parameters are better\" is being dismantled by the \"post-train + distillation + RL\" combination punch. MiniCPM5-1B proves that as long as the post-training flow is solid, 1B can also take on production-grade Agent tasks, and local 1B may no longer be a \"make-do\" backup, but the first choice for low-latency, high-privacy scenarios.","minicpm5-1b-openbmb-17-9-intelligence","2026-06-06T07:00:00Z","2026-06-06T07:13:16.410904Z","2026-08-19T02:08:40.142862Z",true,"agent",164,[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},"0565190a-0bcd-492f-934f-0ad2ab32f485","70万参数2.8MB填一张表:Cua开源CUA-S1,单次前向替代23轮LLM","cua-s1-forms-system-one-model","2026-09-20T13:11:48+00:00",{"id":63,"title":64,"news_slug":65,"published_at":66},"13378d5e-2440-496d-8c3c-7d36858e641d","不聊天的端侧基座:Needle 3 用 8-29MB 在微控制器上跑工具调用","needle-3-tiny-tool-calling-model","2026-09-19T13:09:46+00:00",{"id":68,"title":69,"news_slug":70,"published_at":71},"c99751d5-418e-49d5-99d3-e43b84c80ec7","IBM与NASA开源月球基础模型:Lunar Foundation Model","nasa-ibm-lunar-foundation-model-sombench","2026-09-19T09:30:00+00:00",{"id":73,"title":74,"news_slug":75,"published_at":76},"d056f67b-7e0d-4e44-8d39-e31ea50deeae","Bonsai 2 27B 三元压缩:Qwen3.8 压到 5.9 GB,benchmark 留存 98.2%","bonsai-2-27b-ternary-qwen3-8-compression","2026-09-17T15:47:00+00:00",{"id":78,"title":79,"news_slug":80,"published_at":81},"4c4a2a9e-f69b-4985-bd42-97ab2ef4e2ac","Spark-X2.5-4B 开源:4B 跑 1M 上下文,22 项基准打 9B 级 Qwen3.5","spark-x2-5-4b-apache-open-source","2026-09-16T01:30:00+00:00",{"id":83,"title":84,"news_slug":85,"published_at":86},"dcbb1042-4fef-4a6a-bd7a-0e11b4d97c10","TRL v1.13.0:一处 dtype 修复拿回 21.6% kernel 时间,PPOTrainer 退役","trl-v1-13-tensor-core-lmhead-ppotrainer","2026-09-12T21:05:00+00:00"]