[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-huawei-modelarts-next-agent-native-base":3,"news-related-54b3961a-17a4-4bcf-af68-e4c873eeb8ca":36},{"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},"54b3961a-17a4-4bcf-af68-e4c873eeb8ca","华为云 ModelArts Next：从训练平台走向「智能体原生」的训推底座","在 6 月 5 日开幕的华为云 INSPIRE 2026 创想者大会上，华为云正式发布新一代模型训推平台 ModelArts Next，把「RL 服务、机密推理、模型路由、模型矩阵」作为四大核心能力一次性打包亮相。从命名就可以看出华为云对 Agent 时代的判断：模型不再是单点工具，而是企业 AI 智能体体系的「中枢」。\n\n四大能力中，最具差异化的是 **RLaaS（RL-as-a-Service）** 和 **MaaS 模型路由**。前者把强化学习从大厂实验室能力下沉成可调用的云服务，配合可视化轨迹与训练一致性保障，让车企、金融等强决策场景也能跑得起 RL 微调；后者则依据请求特征动态挑选最优模型，目前已聚合 DeepSeek、Kimi、GLM 等 15+ 款 SOTA 模型，**调度精准率超过 95%，调用成本平均下降 20%**。对 Agent 业务方来说，这意味着不必再为每条 prompt 人工选模型，云侧帮你在性能与成本之间做实时 trade-off。\n\n「**机密推理**」则是面向政企客户的硬通货。模型权重和推理数据全程在可信执行环境中运转，硬件级隔离 + 模型防泄漏机制，把私有化部署的合规门槛降到了「一键」。结合华为自研的 Pangu 行业大模型与 DeepSeek、Kimi、GLM 的 Day-0 接入，企业可以在保留自身数据主权的前提下，调用前沿开源\u002F闭源 SOTA 模型做端到端 Agent 编排。\n\n落地的成绩单同样亮眼：基于 ModelArts 的交通预测精度提升 9.91%，天气预测准确率达 84%，并已沉淀 20+ 行业细分场景。可以预见，当大模型竞争从「拼参数」走向「拼工程化能力」，云厂商的价值锚点正在从「卖 Token」转向「卖训推编排与场景编排」——而 ModelArts Next 显然在押注后者。\n\n**一句话总结：Agent 时代的云战争，已经打到了「训推底座」的层级。**","https:\u002F\u002Fwww.huaweicloud.com\u002Fabout\u002Finspire.html","2a222783-7ba6-412b-9394-951bd06357a4",[10,14,17,20],{"id":11,"name":12,"slug":12,"description":13,"color":13},"6ad31a14-c0da-42df-81fd-564281f768db","agentic-ai",null,{"id":15,"name":16,"slug":16,"description":13,"color":13},"40269b40-7942-4650-9672-ed2e6524d37a","ai-technology",{"id":18,"name":19,"slug":19,"description":13,"color":13},"0a93ec8e-ea39-4693-81de-563ca8c173f7","inference",{"id":21,"name":22,"slug":22,"description":13,"color":13},"01598627-1ea6-4b27-a5d8-874971571a71","llm",[24],{"id":25,"lang":26,"title":27,"summary":28,"content":13},"8147dcc9-af2a-4128-9cf8-f2fe43e9dd73","en","Huawei ModelArts Next: agent-native training and serving base","On June 5 Huawei Cloud released ModelArts Next, positioning the platform from a \"training platform\" to an \"Agent-native\" training-inference base. The new version integrates Agent Runtime, Memory Service, and TokenHub, forming a complete \"model + Agent\" co-design stack. The official numbers show that the Agent Runtime frees 70% of idle compute, the Memory service cuts long-task token usage by 60%, and the TokenHub compute utilization rate is up 40%.","huawei-modelarts-next-agent-native-base","2026-06-05T07:00:00Z","2026-06-05T07:08:58.150471Z","2026-08-19T02:08:40.142862Z",true,"agent",162,{"items":37},[38,43,48,53,58,63],{"id":39,"title":40,"news_slug":41,"published_at":42},"9b47c85f-7b67-4e33-be69-b98fc50af87a","Prime Intellect 押注「递归语言模型」RLM：让 LLM 主动管理自己的上下文","prime-intellect-rlm-recursive-language-model","2026-06-19T14:30:00+00:00",{"id":44,"title":45,"news_slug":46,"published_at":47},"6242571f-c635-4794-b19a-ba08eac4f6d1","NVIDIA 发布全球首款面向 Agent 时代的 CPU：Vera 已送抵 Anthropic、OpenAI、SpaceXAI","nvidia-vera-cpu-agent-anthropic-openai","2026-05-20T01:30:00+00:00",{"id":49,"title":50,"news_slug":51,"published_at":52},"36055e5f-136f-497d-8763-3ed6609f59ff","Meta Muse Glimmer 30B 本地落地:Apache 2.0 的开源智能体,把 Agent 装进 24GB 显存","meta-muse-glimmer-30b-local-agent-apache2-r2","2026-08-19T03:00:00+00:00",{"id":54,"title":55,"news_slug":56,"published_at":57},"5a90a793-8ec1-4b3a-9691-edef5ffe8535","AI「思想病毒」实证:Anthropic 与 EPFL 让恶意想法在 Agent 间自我复制,免疫只需一段警告","mind-viruses-multi-agent-llm","2026-08-18T13:30:00+00:00",{"id":59,"title":60,"news_slug":61,"published_at":62},"777afb24-262f-45cc-961f-d5d49ad42883","AgentOPSD 用递归贝叶斯信念破解多轮 Agent 强化学习的信用分配：清华\u002F浙大\u002F美团让 GRPO 学会看哪个 turn 决定胜负","agentopsd-recursive-belief-credit-assignment","2026-08-07T02:00:00+00:00",{"id":64,"title":65,"news_slug":66,"published_at":67},"3c6fcf46-f5bb-4136-931c-69cd64216e12","Skill-Use 基准揭示 Agent 短板：会做任务，不等于会用 Skill","skill-use-agent-harness-benchmark","2026-08-06T08:00:00+00:00"]