[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-glm-5-3-weights-huggingface-release":3,"news-related-453ce9a1-5d55-4981-b44d-c261b8051724":38},{"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},"453ce9a1-5d55-4981-b44d-c261b8051724","GLM-5.3 753B 权重上架 HuggingFace,智谱兑现两周开源承诺","智谱 8 月 14 日发布 GLM-5.3 时承诺两周内公开权重，如今 753B 参数的模型已上架 HuggingFace。它沿用 GLM-5.2 底座、全部增益来自后训练，官方称其为最强开源编码模型，CyberGym 84.5 领先，还带出 2436 个真实漏洞的披露台账。","8 月 14 日，智谱在 GLM-5.3 的发布文里写下一句承诺：\"权重将在两周后发布，待安全评估与加固完成。\"两周之期刚到，`zai-org\u002FGLM-5.3` 就出现在了 Hugging Face 上——753B 参数、BF16 与 F8_E4M3 精度可选，模型 collection 显示一天前刚更新。承诺没有跳票（权重地址：https:\u002F\u002Fhuggingface.co\u002Fzai-org\u002FGLM-5.3 ）。\n\n## 一场没有换底座的升级\n\nGLM-5.3 最有意思的技术选择是：它与 GLM-5.2 用的是同一个 base model，所有提升全部来自 post-training。官方原话是\"Scaling post-training is all we did\"。\n\n在预训练边际收益被反复讨论的当下，这是一个相当激进的实验：底座不动，靠环境扩展加 RL 把同一个模型再推一档。结果体现在编码与智能体基准上——Terminal Bench 3.0 从 4.6 跳到 28.3，DeepSWE v1.1 从 46.2 到 66.9，Agents' Last Exam 从 23.8 到 28.5；官方口径称其为\"最强的开源权重编码模型\"，自家 Z.ai Code Bench 上相对 GLM-5.2 提升约 50%。以上均为厂商自报数字，未独立复现。\n\n支撑这场后训练的是环境工程：任务不再是\"编码练习题\"，而是数天量级的真实工程工作——模型拿到与工程师相同的工作环境，可以访问计算集群、存储系统、内部文档、代码库和实验结果，要求端到端诊断瓶颈、跑实验、交付可验证的加速。为此智谱建了端到端合成训练环境的 pipeline，judge agent 先尝试每个任务验证它确实可解，再产出足够可靠、可直接训练的二值奖励。开源的 slime 框架承载了整套 RL 扩展。\n\n## 意外长出来的安全能力\n\n发布文里最耐读的部分是\"涌现的网络安全能力\"。智谱本来只是往训练里加入了漏洞发现数据与环境，没料到能力随规模持续窜升：GLM-5.3 不再只识别孤立缺陷，而是能跨利用阶段推理、组织完整的利用链。\n\n数字上（厂商自报）：CyberGym 84.5，官方称其为该基准最佳结果，超过 GPT-5.6 Sol 的 83.6；ExploitBench 54.4，是 GLM-5.2（24.4）的两倍多。但在更贴近完整利用链的 ExploitGym 上，GLM-5.3 完成 105\u002F130 个任务（2h\u002F6h 预算），与闭源前沿模型仍有明显差距——官方自己承认\"能力增长最快的区域，恰恰是落后最远的区域\"。\n\n真实世界的验证更硬核：与多家国内安全团队合作，模型在 269 个项目里识别出 2,436 个漏洞，其中 1,097 个中高危；最老的一个 1981 年就引入了，平均潜伏 26.6 年。目前 53 个已公开披露，其余在 embargo 流程中，全部记录在公开的 Z.ai Security Disclosure Ledger（cvd.z.ai）上。\n\n## 权重落地之后\n\n对开发者，这次上架意味着三件事。其一，模型卡列出了 SGLang、vLLM、TokenSpeed、Transformers、KTransformers、Unsloth 的本地部署支持，还有 Ascend NPU 平台路线——国产算力栈被当作一等公民对待。其二，Hugging Face 页面显示已有 2 个基于它的量化版本跟进，部署门槛会快速下探。其三，注意它与 GLM-5.3-Flash 的分工：Flash 是 320B 总参\u002F18B 激活、MIT 许可的原生多模态模型，本体 753B 则是纯文本旗舰，推理预算差一个量级，按需选型别拿错。\n\n一点观察：当\"换底座\"的增益越来越贵，\"同底座 + 后训练扩展\"给出了另一条性价比曲线。GLM-5.3 用编码基准上 50% 的提升证明了这条路能走通，而权重公开让所有人都能亲手验证它到底走了多远——这或许才是开源承诺真正的价值。","https:\u002F\u002Fhuggingface.co\u002Fzai-org\u002FGLM-5.3","df9ef325-77c5-4e95-9c03-f6cf5b150ef0",[11,15,18,21],{"id":12,"name":13,"slug":13,"description":14,"color":14},"a8002d98-9df1-4ab9-94d4-a7625af634c4","china-ai",null,{"id":16,"name":17,"slug":17,"description":14,"color":14},"01598627-1ea6-4b27-a5d8-874971571a71","llm",{"id":19,"name":20,"slug":20,"description":14,"color":14},"7e89b5cc-57db-4f37-bc6d-28919a73931c","model-release",{"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},"15028894-42a8-47ff-8a1a-4282af366dad","en","GLM-5.3 753B Weights Land on Hugging Face as Z.ai Keeps Its Open-Source Promise","Z.ai promised open weights two weeks after GLM-5.3's August 14 launch, and the 753B model has now landed on Hugging Face. It reuses the GLM-5.2 base with all gains from post-training, is positioned as the strongest open-weights coding model with a leading CyberGym 84.5, and ships with a disclosure ledger covering 2,436 real vulnerabilities.","On August 14, Z.ai's GLM-5.3 launch post carried a promise: \"We will release the weights in two weeks after launch, once safety evaluation and hardening are complete.\" The two-week clock has just run out, and `zai-org\u002FGLM-5.3` is now live on Hugging Face — 753B parameters, available in BF16 and F8_E4M3 precision, with the model collection updated one day ago. The promise was kept (weights at: https:\u002F\u002Fhuggingface.co\u002Fzai-org\u002FGLM-5.3 ).\n\n## An Upgrade Without a New Base Model\n\nThe most interesting technical choice in GLM-5.3 is that it shares the same base model as GLM-5.2 — every gain comes from post-training. The official phrasing: \"Scaling post-training is all we did.\"\n\nWith pretraining returns under constant debate, this is a bold experiment: freeze the base, and push the same model up another tier purely through environment scaling plus RL. The results show up on coding and agentic benchmarks — Terminal Bench 3.0 jumps from 4.6 to 28.3, DeepSWE v1.1 from 46.2 to 66.9, and Agents' Last Exam from 23.8 to 28.5. The company positions it as \"the most capable open-weights model for coding,\" with roughly a 50% improvement over GLM-5.2 on its in-house Z.ai Code Bench. All of these are vendor-reported numbers, not independently reproduced.\n\nWhat sustains this post-training run is environment engineering: tasks are no longer \"coding exercises\" but days-long units of real engineering work — the model gets the same working environment as an engineer, with access to compute clusters, storage systems, internal documentation, codebases, and experiment results, and must diagnose bottlenecks end to end, run experiments, and deliver a verifiable speedup. To scale this, Z.ai built pipelines that synthesize training environments end to end; a judge agent attempts each task first to verify it is actually solvable, then produces binary rewards reliable enough to train on directly. The open-source slime framework carries the whole RL scaling effort.\n\n## A Security Capability Nobody Asked For\n\nThe most striking section of the launch post is the \"emergent cyber capability.\" Z.ai only added vulnerability-discovery data and environments to the training mix, and did not expect the capability to keep climbing as training scaled: GLM-5.3 no longer just spots isolated flaws — it reasons across multiple stages of exploitation and assembles coherent attack chains.\n\nIn numbers (vendor-reported): CyberGym 84.5, which the company calls the best result on that benchmark, ahead of GPT-5.6 Sol at 83.6; ExploitBench 54.4, more than double GLM-5.2's 24.4. But on ExploitGym, which sits closest to full exploitation chains, GLM-5.3 completes 105 tasks within two hours and 130 within six — still clearly behind the closed frontier, and the company itself admits \"capability is growing fastest exactly where we are furthest behind.\"\n\nThe real-world validation is even harder-edged: working with several security teams in China, the model identified 2,436 vulnerabilities across 269 projects, 1,097 of them medium-to-high severity; the oldest dated back to 1981, and the average vulnerability had lurked for 26.6 years. So far 53 findings are publicly disclosed while the rest move through embargo, all tracked in the public Z.ai Security Disclosure Ledger (cvd.z.ai).\n\n## After the Weights Landed\n\nFor developers, this listing means three things. First, the model card lists local-serving support for SGLang, vLLM, TokenSpeed, Transformers, KTransformers, and Unsloth, plus a route for the Ascend NPU platform — the domestic compute stack is treated as a first-class citizen. Second, the Hugging Face page already shows 2 community quantizations building on it, so deployment barriers will fall quickly. Third, note the division of labor with GLM-5.3-Flash: Flash is the 320B-total\u002F18B-activated natively multimodal model under MIT, while the full 753B model is the text-only flagship — an order of magnitude apart in inference budget, so pick deliberately.\n\nOne observation: as gains from \"swapping the base model\" get more expensive, \"same base + post-training scaling\" offers an alternative cost-performance curve. GLM-5.3 demonstrated the path is viable with a 50% coding-benchmark lift, and the open weights let everyone verify with their own hands how far it actually got — perhaps that is the real value of an open-source promise.","glm-5-3-weights-huggingface-release","2026-08-28T15:15:00Z","2026-08-28T15:08:00.611666Z","2026-08-28T15:08:00.611676Z",true,"agent",124,{"items":39},[40,45,50,55,60,65],{"id":41,"title":42,"news_slug":43,"published_at":44},"f6e4aab0-7693-4c2c-bb66-c1641fc2cc3e","Ox Alpha 谜底揭晓:智谱 GLM-5.3-Flash,MIT 开源 320B MoE","ox-alpha-glm-5-3-flash-reveal","2026-08-27T13:30:00+00:00",{"id":46,"title":47,"news_slug":48,"published_at":49},"804ab59a-a8d6-4b61-bf74-8f6f2bdae83c","智谱把 Flash 做成一件正经事:一次说清 GLM-5.3-Flash 的架构和 benchmark 真相","glm-5-3-flash-hybrid-attention-architecture","2026-08-27T08:00:00+00:00",{"id":51,"title":52,"news_slug":53,"published_at":54},"b0183d10-bcfd-44ed-a178-a2c813f10b69","国家超算互联网AI社区上线Kimi K3:2.8万亿参数MoE一键调用,开源大模型有了国产算力底座","kimi-k3-cnsc-internet-launch","2026-07-28T09:30:00+00:00",{"id":56,"title":57,"news_slug":58,"published_at":59},"3d8b9b1a-e038-466f-9b6b-304f911e35a7","Kimi K3 开源三件套 MoonEP\u002FFlashKDA\u002FAgentEnv:Moonshot 把 2.8T MoE 训练栈完整交底","kimi-k3-moonep-flashkda-agentenv","2026-07-28T04:30:00+00:00",{"id":61,"title":62,"news_slug":63,"published_at":64},"a151db0c-d832-4df2-ac03-2d4e58b26e99","Kimi K3 跑通 MiniTriton:Moonshot 让 LLM 第一次从零编译出自己的 GPU 编译器","kimi-k3-minitriton-gpu-compiler","2026-07-26T14:00:00+00:00",{"id":66,"title":67,"news_slug":68,"published_at":69},"ebb562ad-9213-4db8-a29e-28dfba8df066","Kimi K3 上线:Moonshot 用 2.8 万亿参数与 KDA 线性注意力把开源带回牌桌","kimi-k3-launch-2-8t","2026-07-16T20:01:00+00:00"]