央广网上海9月12日消息（记者 殷雨婷）9月12日，蚂蚁数科在2025Inclusion·外滩大会论坛上宣布，将推出按"效果付费"的新型商业模式。该模式支持 企业客户根据大模型应用的实际效果（如业务增长或成本节省）来付费，而非传统的项目制或订阅制模式。 随着大模型从技术层向场景层渗透，AI已成为业务成果的加速器，"企业级AI服务的衡量标准，应从'为技术付费'向'为效果付费'转变，一些以结果为重 的产品，客户可以选择为可见的效果买单这种付费方式。"蚂蚁数科副总裁余滨表示，目前蚂蚁数科已经与部分战略级客户达成基于关键业务效果的合作模 式，将逐步向全行业推广。 蚂蚁数科副总裁余滨在2025外滩大会演讲。受访者供图 "按效付费"模式有望显著降低企业应用大模型的门槛和风险，结合蚂蚁数科的全栈企业级AI服务落地方案体系，尤其有助于中小企业更轻松地接入AI能 力，加速大模型在金融、能源、零售等千行百业的规模化落地。 余滨表示，蚂蚁数科基于行业实践，已经形成大模型落地产业的"ACE"（Align-Construct-Evaluate）三步方法论：首先是与客户进行价值定位，一同找 到可衡量的、最有价值的业务目标。第二步，根据业 ...

据了解，"按效付费"模式将显著降低企业应用大模型的门槛和风险，结合蚂蚁数科的全栈企业级AI服务 落地方案体系，企业无需投入高昂的基础设施成本，即可收获AI带来的业务价值，尤其将助力中小企 业更轻松地接入AI能力，加速大模型在金融、能源、零售等千行百业的规模化落地。 余滨表示，蚂蚁数科基于行业实践，已经形成大模型落地产业的"ACE"（Align-Construct-Evaluate）三 步方法论：首先是与客户进行价值定位，一同找到可衡量的、最有价值的业务目标。第二步，根据业务 目标，为客户构建一整套从模型到应用的解决方案与实施路径；第三步，通过持续联合运营，确保业务 效果最终达成。（文猛） 责任编辑：江钰涵 新浪科技讯 9月12日上午消息，蚂蚁数科在2025Inclusion·外滩大会论坛上宣布，将推出按"效果付费"的 新型商业模式。该模式支持企业客户根据大模型应用的实际效果（如业务增长或成本节省）来付费，而 非传统的项目制或订阅制模式。 随着大模型从技术层向场景层渗透，AI已经不再是单点技术工具，而正成为业务成果的加速器，"企业 级AI服务的衡量标准，应从'为技术付费'向'为效果付费'转变，一些以结果为重的产 ...

Core Insights - Baidu Smart Cloud has partnered with 65% of central enterprises to advance the implementation of large model industries, leveraging its comprehensive AI capabilities to create replicable benchmark cases across various business scenarios [1][3] Group 1: AI Technology and Industry Applications - The 2025 Smart Economy Forum highlighted AI's role in real-world applications, showcasing how large models empower various industries [3] - Baidu launched the Qianfan Huijin financial model, designed specifically for the financial sector, featuring two versions (8B and 70B parameters) and supporting up to 32K context input [4][8] - The Qianfan platform serves as a leading enterprise-level "intelligent agent factory," facilitating the rapid development of industry-specific intelligent agents across high-value sectors like energy, transportation, and healthcare [5][7] Group 2: Model Performance and Capabilities - The Qianfan Huijin financial model outperformed general models in financial benchmarks, demonstrating superior industry knowledge and reasoning capabilities [8] - Baidu's new flagship models, Wenxin 4.5 Turbo and X1 Turbo, significantly enhance multi-modal understanding and reasoning abilities, with improvements exceeding 30% in multi-modal comprehension [9] Group 3: Infrastructure and Support - Baidu's self-developed GPU computing platform, Baijiao, supports various mainstream models and optimizes overall computing costs to half of the market price, ensuring reliable support for enterprise model training [10] - The collaboration with Changan Automobile has established a computing center that provides real-time inference support for all models, achieving a total computing power exceeding 1000 PFLOPs [10]