广告
加载中

阿里文征:面向Agent全面升级模型服务、Agent服务、AI应用

龚作仁 2026-09-23 19:26
龚作仁 2026/09/23 19:26

邦小白快读

EN
全文速览

千问AI平台全面升级,核心是让AI从回答问题升级为直接帮你完成任务,并已上线多项好用的新功能。

1. 新推出API优速模式,把请求处理速度提升1.5至2倍,开发者换一下模型ID就能用,普通人用相关工具也会感觉更快。

2. 发布Agent Studio,能自动选择模型、24小时托管运行,还支持接入企业知识库和外部软件,相当于有了一个帮你搭建专属智能助手的工厂。

3. 推出Token Plan套餐,一份订阅就能用Qwen、Kimi、GLM、DeepSeek等多个模型,还能和千问App、千问办公共用额度,并兼容Cursor、Codex等常用编程工具。

4. 生态连接更方便:One Key MCP用一个API Key就能连接高德、飞猪、1688等100多项服务,智能助手能自行找到并调用工具。

5. 模型启动时间从1200秒缩短到70秒,成本最高节省95%,意味着AI服务更便宜、响应更快,手机和汽车上的智能助手也会更聪明,比如比亚迪能听懂模糊指令,荣耀手机能主动服务。

品牌商可从千问AI平台升级中获取品牌营销、产品体验和用户洞察方面的干货:

1. 平台新增Agent Studio和行业AI解决方案,品牌可基于自身知识数据和外部服务,搭建客服、营销等智能体,提升用户互动和转化,例如伶鹊2.0已服务20个行业、5000余家企业客户和超10万坐席。

2. 产品研发方面,荣耀采用Qwen Intelligence方案,以端云协同实现多模态理解和长链路任务规划,推动AI手机从被动响应到主动服务;比亚迪基于千问AI座舱方案打造超级智能体,能理解和执行模糊、多重指令,这些都是品牌打造差异化智能产品的参考方向。

3. 用户行为观察上,文章指出AI需求不再需要证明,Token用量持续增长,过去一年阿里云MaaS客户数增长六倍,说明越来越多的企业把智能转化为结果,品牌商应重视Agent进入核心流程的趋势,思考如何通过AI交付实际业务价值,而不是只调用API。

4. 成本与定价方面,Token Plan一份订阅覆盖多模型,API优速模式提升TPS且成本最高节省95%,品牌商可以更低成本尝试多种模型,降低营销和客服AI化的门槛。

卖家能从这次千问AI平台升级中看到新的增长市场和商业机会,主要有以下几点:

1. 增长市场:过去一年阿里云MaaS平台客户数增长六倍,Agent开始进入企业核心流程,说明企业级AI服务市场正在快速放大,卖家可围绕Agent开发、运维或提供行业数据、服务接入等方式切入。

2. 消费需求变化:客户不再满足于消耗Token,而是要交付结果,单任务价值、Token生产效率和单Token能力成为衡量AI价值的核心,卖家应转向能直接带来业务结果的智能服务。

3. 机会与工具:平台推出Agent Studio,提供50多个原子API、自动路由选模、24小时托管和自进化引擎,卖家可以低成本搭建行业Agent;One Key MCP支持一个API Key连接高德、飞猪、1688等100多项服务,为电商卖家打通更多流量和服务入口。

4. 应对措施与合作方式:相关模型服务和Agent能力已上线,商家可通过API集成或订阅Token Plan;Token Plan覆盖多个主流模型,并兼容主流工具,卖家可借助这些能力升级店铺运营、客户服务和营销投放。

5. 风险提示:AI仍处于早期,用户覆盖和场景渗透有增长空间,但也意味着市场变化快,需关注成本治理和模型性能,建议选择有SLA、监控告警和成本治理能力的平台服务。

工厂可从千问AI平台升级中获得生产数字化和电商拓展的直接启发:

1. 生产与工艺智能化:泛海集团已基于千问搭建工业AI平台和数控工艺Agent,把工程师经验和零件案例沉淀到模型中,工厂可以参考此做法,用Agent保存老师傅知识,辅助工艺设计与生产决策。

2. 模型服务基础能力:平台提供99.9%生产级SLA、十亿级单客户峰值TPM、CMaaS机密推理,以及FlashBoot将模型弹性启动时间缩至70秒、1分钟拉起1万个Pods,可满足工厂对稳定性和数据安全的要求。

3. 低门槛开发:Agent Studio支持企业内部知识数据和外部软件接入,提供50多个原子API,工厂无需从零训练模型,按需组合即可构建质检、排产、设备运维等工业Agent。

4. 电商与获客:万小智3.0已从建站延展至获客和经营,推动“Vibe Coding”走向“Vibe Business”,工厂可利用这些工具搭建线上渠道并更高效获取客户。

5. 算力保障:阿里云计划到2032年算力规模超过20GW,持续投入全栈AI,为工业AI应用提供长期基础设施支撑。

对服务商而言,这篇文章披露了千问AI平台的最新技术能力、行业趋势和可服务的客户痛点:

1. 行业趋势:AI价值衡量方式正从消耗Token转向交付结果,智能价值密度由单任务价值、Token生产效率和单Token能力共同决定;Agent开始进入企业核心流程,客户需要的不是模型API,而是覆盖模型供给、运行、构建与托管的完整体系。

2. 新技术与解决方案:Agent Studio提供自动路由选模、24小时托管、50多个原子API、自进化引擎(运行观测、AI评测器、AI优化器、自动验证);One Key MCP用一个API Key连接100多项生态服务,能显著降低Agent开发集成成本。

3. 性能与成本优化:FlashBoot将模型弹性启动时间由1200秒降至70秒,1分钟拉起1万个Pods;首Token延迟降低38%,Prompt Caching成本最高节省95%;API优速模式可将TPS提升1.5至2倍。

4. 客户痛点应对:模型服务提供99.9%生产级SLA、十亿级TPM、CMaaS机密推理、监控告警和成本治理,并有Token Plan、按量付费、PTU、DTU等灵活模式,服务商可据此为客户设计差异化方案。

5. 实践参考:文章列举泛海工业AI、网易游戏、荣耀与比亚迪智能终端等案例,服务商可借鉴这些落地路径。

平台商可从千问AI平台的升级中看到平台建设、生态运营和风险管理的具体做法:

1. 平台定位:平台的价值不是定义所有应用,而是让更多应用生长出来;千问AI平台以推动Agent进入生产为核心,在模型服务基础上新增Agent服务、行业AI解决方案,构建模型-工具-应用的全栈体系。

2. 招商与合作:通过One Key MCP用一个API Key连接高德、飞猪、1688及金融、法律等领域100多项生态服务,降低服务商接入门槛,吸引更多开发者;Agent Studio提供全栈Agent托管、自动路由和自进化能力,可作为平台吸引开发者的核心卖点。

3. 运营管理:平台提供99.9%生产级SLA、监控告警和成本治理,帮助客户管理资源;推出API优速模式(TPS提升1.5-2倍)、Token Plan(一份订阅覆盖多模型并兼容主流工具),以灵活计费模式增强客户粘性。

4. 风险与合规:面向大型企业及金融、政企客户提供独立吞吐DTU和CMaaS机密推理服务,满足数据安全与合规要求。平台应重视安全和性能保障,并持续投入全栈AI,2032年算力超20GW,打造长期竞争力。

5. 风向把握:AI需求仍处早期,用户覆盖、场景渗透均有空间,平台可抓住智能交付趋势,推动Agent进入企业核心流程。

本文反映了AI产业从模型调用到智能交付的关键转向,值得关注的产业新动向、技术体系与商业模式:

1. 产业新动向:阿里云MaaS客户数一年增长六倍,Agent开始进入订单、代码、内容和业务流程等企业核心流程,智能正从被调用走向被交付;但AI仍处早期,用户覆盖、场景渗透和算力规模有巨大增长空间。

2. 技术与基础设施:阿里云预计到2032年算力规模超20GW,支撑芯片、AI基础设施、千问大模型、模型服务、Agent工具和AI原生应用的全栈体系;FlashBoot将模型弹性启动时间从1200秒降至70秒,1分钟拉起1万个Pods,首Token延迟降38%,Prompt Caching成本省95%,这些是研究AI基础设施演进的重要数据。

3. 商业模式:平台推出Token Plan订阅(覆盖多模型并兼容主流工具)、按量付费、API优速模式、PTU(夜间8小时预留)和DTU(独占定制)等分层计费模式,反映AI服务从单一API向结果交付演进的价值衡量方式变化。

4. 平台生态策略:Agent Studio提供自动路由、50多个原子API、自进化引擎;One Key MCP用一个API Key连接100多项生态服务,体现平台开放能力与降低集成门槛的设计思路。

5. 落地案例与启示:泛海工业AI、网易游戏、荣耀AI手机、比亚迪AI座舱等案例展示了Agent在不同行业的应用路径,为研究智能体生产化提供了样本;但文章未涉及政策法规,产业界需进一步关注数据安全与行业治理问题。

返回默认

声明:快读内容全程由AI生成,请注意甄别信息。如您发现问题,请发送邮件至 run@ebrun.com 。

我是 品牌商 卖家 工厂 服务商 平台商 研究者 帮我再读一遍。

Quick Summary

The Qianwen AI platform has undergone a comprehensive upgrade, with the core shift from answering questions to directly completing tasks for users, along with several new practical features.

1. A new API Express mode boosts request processing speed by 1.5 to 2 times; developers can activate it simply by changing the model ID, and ordinary users will also notice faster response times in related tools.

2. Agent Studio is now available, which automatically selects models, supports 24/7 hosted operation, and can integrate with enterprise knowledge bases and external software — essentially a factory for building your own customized intelligent assistant.

3. A new Token Plan subscription lets users access multiple models including Qwen, Kimi, GLM, and DeepSeek with one plan, while sharing quota with the Qianwen App and Qianwen Office; it is also compatible with popular coding tools such as Cursor and Codex.

4. Ecosystem connectivity is more convenient: One Key MCP uses a single API key to connect to over 100 services including AMAP, Fliggy, and 1688, allowing intelligent assistants to discover and invoke tools on their own.

5. Model startup time has been reduced from 1,200 seconds to 70 seconds, with cost savings of up to 95%. This means AI services are cheaper and faster, and smart assistants in phones and cars will become more capable — for instance, BYD can understand vague commands, and Honor phones can proactively offer services.

Brand owners can extract actionable insights from the Qianwen AI platform upgrade in brand marketing, product experience, and user understanding:

1. With the new Agent Studio and industry-specific AI solutions, brands can build intelligent agents for customer service and marketing based on their own knowledge data and external services, improving user engagement and conversion. For example, Lingque 2.0 already serves 20 industries, over 5,000 enterprise clients, and more than 100,000 seats.

2. On product R&D, Honor has adopted the Qwen Intelligence solution, using cloud-device collaboration to achieve multimodal understanding and long-horizon task planning, pushing AI phones from passive response to proactive service. BYD has built a super agent based on Qianwen's AI cockpit solution, capable of understanding and executing vague, multi-part commands. These serve as reference directions for brands developing differentiated intelligent products.

3. On user behavior, the article notes that AI demand no longer needs to be proven: token consumption continues to grow, and Alibaba Cloud's MaaS customer base has increased sixfold over the past year. This shows more enterprises are converting intelligence into outcomes. Brands should take seriously the trend of agents entering core business processes and think about how to deliver real business value through AI rather than merely calling APIs.

4. On cost and pricing, a Token Plan covers multiple models with a single subscription, while API Express mode improves TPS and cuts costs by up to 95%. Brands can experiment with more models at lower cost, lowering the barrier to AI-driven marketing and customer service.

Sellers can identify new growth markets and business opportunities from this Qianwen AI platform upgrade:

1. Growth market: Over the past year, Alibaba Cloud's MaaS customer base has grown sixfold, and agents are beginning to enter core enterprise processes. The enterprise AI services market is expanding rapidly, and sellers can enter by developing or operating agents, or by providing industry data and service integration.

2. Changing consumer demand: Customers are no longer satisfied with consuming tokens; they want delivered results. Per-task value, token production efficiency, and per-token capability have become the core metrics of AI value. Sellers should shift toward intelligent services that directly deliver business outcomes.

3. Opportunities and tools: The platform's Agent Studio provides more than 50 atomic APIs, automatic model routing, 24/7 hosting, and a self-evolving engine, allowing sellers to build industry agents at low cost. One Key MCP connects more than 100 services such as AMAP, Fliggy, and 1688 with a single API key, opening more traffic and service touchpoints for e-commerce sellers.

4. Actions and collaboration: Related model services and agent capabilities are now available; sellers can integrate via APIs or subscribe to Token Plan. The Token Plan covers multiple mainstream models and is compatible with common tools, enabling sellers to upgrade store operations, customer service, and marketing campaigns.

5. Risk warning: AI is still early-stage, with room for user coverage and scenario penetration, but this also means rapid market change. Sellers should pay attention to cost governance and model performance, and choose platforms with SLA, monitoring, alerting, and cost management capabilities.

Factories can gain direct inspiration from the Qianwen AI platform upgrade for production digitalization and e-commerce expansion:

1. Intelligent production and process control: Fanhai Group has built an industrial AI platform and CNC process agent based on Qianwen, embedding engineer experience and part cases into the model. Factories can follow this approach to use agents to preserve veteran workers' knowledge and assist process design and production decisions.

2. Foundational model service capabilities: The platform offers 99.9% production-grade SLA, billion-level peak TPM per customer, CMaaS confidential inference, and FlashBoot reduces model elastic startup time to 70 seconds and can spin up 10,000 Pods in one minute, meeting factories' requirements for stability and data security.

3. Low-barrier development: Agent Studio supports integration with enterprise internal knowledge data and external software, and offers more than 50 atomic APIs. Factories do not need to train models from scratch; they can combine components on demand to build industrial agents for quality inspection, production scheduling, equipment maintenance, and more.

4. E-commerce and customer acquisition: Wanxiaozhi 3.0 has expanded from website building to customer acquisition and operations, advancing "Vibe Coding" to "Vibe Business." Factories can use these tools to build online channels and acquire customers more efficiently.

5. Computing power assurance: Alibaba Cloud plans to reach over 20GW of computing capacity by 2032, with continued investment in full-stack AI, providing long-term infrastructure support for industrial AI applications.

For service providers, this article reveals the latest technical capabilities of the Qianwen AI platform, industry trends, and addressable customer pain points:

1. Industry trend: The measure of AI value is shifting from token consumption to delivered results. Intelligent value density is determined by per-task value, token production efficiency, and per-token capability. Agents are beginning to enter core enterprise processes. Customers no longer need just model APIs; they need a complete system covering model supply, operation, building, and hosting.

2. New technologies and solutions: Agent Studio provides automatic model routing, 24/7 hosting, more than 50 atomic APIs, and a self-evolving engine (runtime observation, AI evaluator, AI optimizer, automatic validation). One Key MCP connects over 100 ecosystem services with a single API key, significantly reducing agent development and integration costs.

3. Performance and cost optimization: FlashBoot reduces model elastic startup time from 1,200 seconds to 70 seconds, and can spin up 10,000 Pods in one minute. Time-to-first-token latency drops by 38%, and Prompt Caching cuts costs by up to 95%. API Express mode can improve TPS by 1.5 to 2 times.

4. Addressing customer pain points: The model service provides 99.9% production-grade SLA, billion-level TPM, CMaaS confidential inference, monitoring, alerting, and cost governance, along with flexible pricing modes such as Token Plan, pay-as-you-go, PTU, and DTU. Service providers can use these to design differentiated solutions for clients.

5. Practical references: The article cites cases such as Fanhai Industrial AI, NetEase Games, Honor and BYD's intelligent devices. Service providers can learn from these implementation paths.

Platform operators can observe concrete practices in platform construction, ecosystem operations, and risk management from the Qianwen AI platform upgrade:

1. Platform positioning: The value of a platform is not to define all applications but to let more applications grow. The Qianwen AI platform focuses on pushing agents into production, adding agent services and industry AI solutions on top of model services, building a full-stack model-tool-application system.

2. Recruiting and collaboration: One Key MCP connects more than 100 ecosystem services across AMAP, Fliggy, 1688, finance, and legal sectors with a single API key, lowering the barrier for service provider integration and attracting more developers. Agent Studio provides full-stack agent hosting, automatic routing, and self-evolution capabilities, which can serve as a core selling point to attract developers.

3. Operations management: The platform offers 99.9% production-grade SLA, monitoring, alerting, and cost governance to help customers manage resources. API Express mode (TPS up 1.5-2 times) and Token Plan (one subscription, multiple models, compatible with mainstream tools) use flexible pricing to strengthen customer stickiness.

4. Risk and compliance: For large enterprises, finance, and government clients, the platform provides dedicated DTU throughput and CMaaS confidential inference services to meet data security and compliance requirements. Platforms should prioritize safety and performance, continue investing in full-stack AI, and achieve over 20GW of computing capacity by 2032 to build long-term competitiveness.

5. Trend awareness: AI demand is still early-stage, with room for user coverage and scenario penetration. Platforms can seize the intelligent delivery trend and push agents into enterprise core processes.

This article reflects a key shift in the AI industry from model invocation to intelligent delivery. It highlights new industry trends, technical systems, and business models worth attention:

1. New industry trend: Alibaba Cloud's MaaS customer base grew sixfold in one year, and agents are starting to enter core enterprise processes such as orders, code, content, and business workflows. Intelligence is moving from being invoked to being delivered. However, AI is still in its early stage, with enormous room for user coverage, scenario penetration, and computing capacity growth.

2. Technology and infrastructure: Alibaba Cloud expects to have over 20GW of computing capacity by 2032, supporting a full-stack system spanning chips, AI infrastructure, Qwen foundation models, model services, agent tools, and AI-native applications. FlashBoot reduces model elastic startup time from 1,200 seconds to 70 seconds, spins up 10,000 Pods in one minute, cuts time-to-first-token latency by 38%, and reduces Prompt Caching costs by 95% — important data for studying AI infrastructure evolution.

3. Business model: The platform introduces Tiered pricing such as Token Plan subscription (covering multiple models and compatible with mainstream tools), pay-as-you-go, API Express mode, PTU (nightly 8-hour reservation), and DTU (dedicated custom throughput). This reflects the shift in value measurement from a single API to outcome delivery.

4. Platform ecosystem strategy: Agent Studio provides automatic routing, more than 50 atomic APIs, and a self-evolving engine. One Key MCP connects more than 100 ecosystem services with a single API key, demonstrating the design approach of open platforms and reducing integration barriers.

5. Implementation cases and implications: Cases such as Fanhai Industrial AI, NetEase Games, Honor AI phones, and BYD AI cockpits show how agents are applied across industries, providing samples for researching agent productionization. However, the article does not address policy and regulation; the industry needs to further focus on data security and governance issues.

Disclaimer: The "Quick Summary" content is entirely generated by AI. Please exercise discretion when interpreting the information. For issues or corrections, please email run@ebrun.com .

I am a Brand Seller Factory Service Provider Marketplace Seller Researcher Read it again.

9月22日,在2026云栖大会MaaS&Agent技术主论坛上,阿里巴巴集团战略副总裁、ATH事业群MaaS业务线总裁文征宣布千问AI平台全面升级:以推动Agent进入生产作为建设核心,在模型服务基础上新增对Agent服务、行业AI解决方案的支持,并推出了API优速模式、Agent Studio、千问AI座舱解决方案等一系列能力,帮助企业将智能转化为生产力。

从消耗Token到交付结果,AI价值的衡量方式正在改变

文征表示,AI仍处于非常早期的阶段,用户覆盖、场景渗透和算力规模均有巨大增长空间。持续增长的Token用量说明,AI需求已不再是一道需要证明的题,关键在于谁能将智能转化为结果。单任务价值、Token生产效率和单Token能力,共同决定“智能价值密度”。

过去一年,阿里云MaaS平台服务的客户数增长六倍。随着订单、代码、内容和业务流程主动发起调用,Agent开始进入企业核心流程,企业需要的不再只是一个模型API,而是一套覆盖模型供给、生产运行、Agent构建与托管的完整体系。智能正在从被调用走向被交付。

需求的持续放大,对算力和基础设施提出了更高要求。阿里将坚定投入全栈AI,预计到2032年,阿里云运营的算力规模将超过20GW,支撑从芯片、AI基础设施,到千问大模型、模型服务、Agent工具和AI原生应用的完整AI体系,加速Agent时代的产业进化。

面向Agent负载持续完善模型服务

在模型供给方面,千问AI平台以“First and Best”原则,支持全球一流的自研和开放模型第一时间上线。针对Agent决策链路长、并发高、对时延敏感等特点,平台通过FlashBoot、UniScheduler等能力调度异构算力,将模型弹性启动时间由1200秒缩短至70秒,可在1分钟内拉起1万个Pods;首Token延迟降低38%,Prompt Caching成本最高节省95%。

平台提供99.9%生产级SLA、十亿级单客户峰值TPM和CMaaS机密推理服务,支持监控、告警和成本治理。本次升级的API优速模式可将TPS提升1.5至2倍,用户切换model ID即可使用;吞吐预留(PTU)新增夜间8小时预留规格;独占吞吐(DTU)则面向大型企业及金融、政企客户,支持定制模型部署与性能定制。

在服务模式上,Token Plan以一份订阅覆盖Qwen、Kimi、GLM、DeepSeek等多类模型,原生打通Qoder、Cursor、Codex、OpenClaw等主流编程工具与Agent框架,并可与千问App、千问办公共享抵扣。结合按量付费、优速模式、吞吐预留与独占吞吐,从个人到企业的各类需求均可在同一平台获得对应供给。

Agent Studio发布,提供企业级全栈Agent服务

从模型调用到业务结果,Agen需要模型匹配、任务运行、上下文接入和服务调用等基础能力。千问AI平台正式发布企业级全栈Agent服务平台Agent Studio,可根据任务自动路由选择模型,提供24小时不间断托管运行,并支持企业内部知识数据和外部软件服务接入。

在Agent开发运行方面,平台提供原子API,50多个原子API覆盖运行环境、长期记忆、工具服务、会话请求和服务端点部署等能力,企业可按需组合并接入自身业务;全新的Agent API即将发布,可将环境、记忆、工具、部署等需逐项调用的配置,一次请求全部搞定。

Agent Studio还将推出Agent自进化引擎,包含运行观测、AI评测器、AI优化器和自动验证等能力,支持在上下文交互中持续调优与进化。

同时,针对Agent生态服务,One Key MCP可用一个API Key连接100多项生态服务,包括高德、飞猪、1688及金融、法律等各领域服务。平台还可自动理解需求、寻找工具并完成调用,减少开发者逐一注册和配置鉴权的工作,

以全栈基础能力,支撑应用与行业落地

在应用和行业侧,千问AI平台通过自身产品实践持续验证技术和体验,并将沉淀能力向生态开放。Meoo 1.0通过CLI和OpenAPI提供应用构建、部署及企业协作能力;伶鹊2.0服务20个行业、5000余家企业客户和超过10万个坐席;万镜一刻以Agent、创作工作台和技能广场服务视频生产,万小智3.0则从建站延展至获客和经营,推动“Vibe Coding”走向“Vibe Business”。

在行业与终端场景,泛海集团基于千问搭建工业AI平台和数控工艺Agent,沉淀工程师经验与零件案例;网易游戏与千问AI平台围绕游戏研发链路开展合作。荣耀基于Qwen Intelligence方案,以端云协同的方式,为荣耀YOYO智能体提供多模态理解、长链路任务规划等能力,推动AI手机从被动响应走向主动服务;比亚迪依托千问AI座舱解决方案打造座舱超级智能体,支持理解和执行模糊、多重指令。

文征认为,平台的价值不是定义所有应用,而是让更多应用生长出来。只有将模型服务、Agent服务、AI原生应用和行业实践连成体系,智能才能从通用能力走向产业生产力。

目前,相关模型服务和Agent能力已在千问AI平台上线。开发者可通过API完成产品集成或订阅Token Plan,在已支持的常用AI工具中直接调用相应模型和服务。

本文首发于 亿邦动力 官方网站

文章来源:Laborer

广告
微信
朋友圈

FAQ回顾

千问AI平台是什么?

千问AI平台是阿里云推出的全栈AI平台,提供模型服务、Agent服务和行业AI解决方案,以推动Agent进入生产为核心,帮助企业将智能转化为生产力。

千问AI平台的Agent Studio有哪些功能?

Agent Studio是千问AI平台的企业级全栈Agent服务平台,支持根据任务自动路由选择模型,提供24小时不间断托管运行,可接入企业内部知识数据和外部软件服务,并提供50多个原子API,支持Agent自进化引擎和One Key MCP连接100多项生态服务。

千问AI平台支持哪些大模型?

千问AI平台通过Token Plan支持Qwen、Kimi、GLM、DeepSeek等多类模型,并原生打通Qoder、Cursor、Codex、OpenClaw等主流编程工具与Agent框架,还可与千问App、千问办公共享抵扣。

千问AI平台在性能优化方面有哪些成果?

通过FlashBoot、UniScheduler,千问AI平台将模型弹性启动时间由1200秒缩短至70秒,可1分钟拉起1万个Pods;首Token延迟降低38%,Prompt Caching成本最高节省95%;API优速模式使TPS提升1.5至2倍。

千问AI平台有哪些企业应用案例?

泛海集团用千问搭建工业AI平台和数控工艺Agent;网易游戏在游戏研发中与千问合作;荣耀基于Qwen Intelligence为YOYO智能体提供多模态理解能力;比亚迪依托千问AI座舱解决方案打造座舱超级智能体。

这么好看,分享一下?

朋友圈 分享

APP内打开

+1
+1
微信好友 朋友圈 新浪微博 QQ空间
关闭
收藏成功
发送
/140 0