[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-faabdff7-bbe7-4fb2-9ae9-94eba2a67855":3},{"id":4,"title":5,"summary":6,"original_url":7,"source_id":8,"tags":9,"published_at":23,"created_at":24,"modified_at":25,"is_published":26,"publish_type":27,"image_url":13,"view_count":28},"faabdff7-bbe7-4fb2-9ae9-94eba2a67855","阶跃Step Edge：0.1秒本地toolcall，端云协同补齐Agent落地的最后拼图","阶跃星辰7月12日发布Step Edge端侧模型全家桶，包含基础模型、Audio、GUI、Gen四款产品，把\"端云协同\"作为完整产品形态打包交付，而不只是单点demo。\n\n端侧模型的关键挑战不是把模型做小，而是在功耗、延迟、隐私三个约束下依然可用。Step Edge把工具调用本地延迟压到0.1秒，配合自研Step Inference NPU引擎：1024 token文本输入延迟4.33秒、768分辨率图像识别5.61秒、30秒语音交互10.7秒，Prefill TPS峰值1395次\u002F秒。车载语音、屏幕操控这类Agent场景可以在终端真正落地，不再依赖云端往返。\n\n更值得注意的是阶跃把模型做成了Pro+Flash+Edge三档分层：云端Pro跑复杂推理，Flash支撑高频工作流，端侧Edge承担本地执行。以整理发票为例，Edge先在本地完成语音识别和图像分类，跨年度账目比对才调用Pro——分层既保效率，也把隐私数据留在终端。\n\n测试成绩上，Step Edge基础模型在16项终端任务指标中以62.92分领跑，GUI模型在OSWorld评测中位居榜首，官方称29项核心指标均位列第一。OSWorld榜单此前由Gemini 3.5 Flash等海外旗舰把持，国产GUI Agent首次进入第一梯队——它意味着Agent操作系统不再是云端专属。\n\n行业层面，端云协同正在成为Agent落地的标准范式：Apple Intelligence靠端侧模型保隐私，Google把Gemini Nano塞进Pixel，Anthropic也提出分层推理思路。阶跃把\"Pro+Flash+Edge\"从概念变成现成产品套件，对手机厂商、车企来说是降低Agent集成门槛的捷径。\n\n下一战场不在模型本身而在硬件协同。Step Inference NPU引擎面向终端芯片做了底层优化，Apple找PrismML压1-bit Qwen也是同一条逻辑的不同表达——谁能把模型、推理引擎、芯片三层打通，谁就能定义端侧Agent的下一个标准。","https:\u002F\u002Fwww.itbear.com.cn\u002Fhtml\u002F2026-07\u002F1441199.html","5f7d17cd-f95b-4a76-be2e-db79144de285",[10,14,17,20],{"id":11,"name":12,"slug":12,"description":13,"color":13},"fca9258a-9430-455a-b95d-b9fae5e373a8","ai-inference",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},"499f4b56-819d-49a3-9609-33e775143b86","multimodal","2026-07-14T00:00:00Z","2026-07-14T00:07:54.798652Z","2026-07-14T00:07:54.798674Z",true,"agent",3]