阿里巴巴集团的数字技术与智能骨干业务阿里云今日在其年度开发者峰会上，宣布推出其最新大型语言模型及AI开发工具，升级基础设施产品，以及启动新的全球开发者支持计划。这些新举措旨在让全球开发者能够以更具成本效益的方式构建创新 AI 应用，进而推动全球生成式 AI （GenAI）社群的蓬勃发展。
阿里云弹性计算及存储产品线负责人陈起鲲表示：「在生成AI时代，阿里云坚持技术创新，致力于为AI 模型、AI应用、数据工程、云平台工程等各类开发者提供全面的技术产品和服务。」
为开发者提供更多基础模型和开发工具
全球开发者现可透过阿里云生成式AI开发平台「Model Studio」的API，使用阿里云最新自研大语言模型家族通义千问（Qwen）系列，包括涵盖70亿到720亿参数的通义千问2.5。此外，视觉理解模型Qwen-VL系列、视觉生成模型Wanx2.1（亦称作「通义万象」），以及音频语言模型Qwen-Audio等多模态AI模型，也可供开发者使用。
阿里云基于Qwen 2.5-coder模型的自研通义灵码亦已向开发者开放。这款AI编程助手具备代码补全与优化、调试辅助、代码片段搜索以及批量生成单元测试等功能，可为开发者带来高效流畅的编程体验，显著提升开发效率与创造力。
除提供更丰富的模型外，Model Studio还为全球开发者提供一系列全新AI开发工具。其中，工作流（Workflow）工具可将复杂任务拆解为多个子任务，令工作流程更易掌控；智能体（ Agent）工具则支持多智能体的协作，以便更好地完成规划与执行任务。此外，还有检索增强生成（Retrieval-Augmented Generation, RAG）工具，它能够借助外部资源提升生成式AI模型的精准度与可靠性；批量推理（Batch Reasoning）工具可同时针对多个提示（Prompt）生成相应结果；自动模型评估（Automated Model Evaluation，AutoEval）工具，以及模型部署和应用观测服务等，均将在本月底前上线。
升级基础设施助力AI发展
为提供强大的算力来推动AI普惠并优化其他关键工作负载，阿里云透露，其第9代企业级弹性计算服务（ECS）实例将于今年4月开始在全球市场推出。与上一代相比，新一代的ECS实例在性能上有显著提升，其中包括运算效率提升20%。此外，透过弹性RDMA（Elastic Remote Direct Memory Access，简称eRDMA）加速网路，其支持高效能运算、搜索推荐和 Redis 数据库的性能可进一步提升达 50%。
此外，阿里云创新的容器计算服务（ACS）自 2025 年 1 月起面向国际客户开放。ACS 旨在利用容器技术优化工作任务部署，整合容器服务与底层云计算资源，大幅降低成本和技术障碍，使开发者能够专注于创新而无需兼顾基础设施管理。
全新GenAI计划激发创造力
为推动创新，阿里云推出了GenAI 赋能计划，基于通义千问模型，专为全球开发者和初创公司建构GenAI 应用提供支持。参与者可获免费云计算额度，并会获邀参加培训工作坊、科技展览和演示日，以及获得产品合作推广机会等。计划旨在帮助开发者和初创公司加速推进GenAI项目，同时建立更广泛的创新生态网络。
全球开发者和客户透过通义千问开发前沿应用
专注研发语音助理等AI产品的日本公司Axcxept，基于Qwen 2.5 LLM 开发了一款名为 EZO 的开放源码、轻量级AI模型。EZO模型在编码、信息抽取、数学、推理、角色扮演和日文书写等领域的表现优于最先进的 (SOTA) 模型。EZO具备低延时特性，性能强大，专为日本的医疗与公共机构等产业量身打造，可确保安全且高效的 AI 应用。
Axcxept Inc.首席执行官Kazuya Hodatsu表示：「通义千问2.5大幅增强了处理日文的能力，与其他模型相比更具竞争优势。透过Axcxept的自研训练流程，我们开发了具备高精准度的日语LLM。」
源自牛津大学的创新型科技公司OxValue.AI亦在使用阿里云通义千问的多模态 AI 模型，提供 AI 驱动的公司估值服务。透过处理和分析与融资、研发和营运相关的文字和语音数据，OxValue 能够为企业客户量身定制精准且极具成本效益的估值方案。
OxValue.AI创始人傅晓岚教授表示：「处理多样化的数据来源对于我们的评估服务至关重要。阿里云的AI模型协助我们显著提高了此流程的质量和效率。与阿里云合作，令我们能够为企业客户提供更大的价值。」
Alibaba Cloud, the digital technology and intelligence backbone of Alibaba Group, has unveiled an expanded suite of large language models and AI development tools, upgraded infrastructure offerings, and new support programs for global developers at its annual developer summit today. These advancements aim to empower developers worldwide to build innovative AI applications more cost-effectively and drive a thriving global generative AI community.
“In the era of generative AI, Alibaba Cloud remains committed to technological innovation, aiming to provide comprehensive technical products and services for global developers in fields such as AI models, AI applications, data engineering, and cloud platform services." Alex Chen, Alibaba Cloud Fellow, Head of Product, Compute and Storage
More Foundation Models and Development Tools for Developers
The latest Qwen models, Alibaba Cloud’s proprietary large language model family, including Qwen2.5 series with sizes ranging from 7billion to 72billion parameters, are now accessible via APIs on its generative AI development platform, Model Studio, for global developers to use. Additionally, multimodal AI models including vision understanding models such as Qwen-VL series, visual generation model Wanx2.1 (also known as Tongyi Wanxiang), and audio language model Qwen-Audio are also available for developers to access.
Developers can also leverage Tongyi Lingma, Alibaba Cloud’s proprietary AI coding assistant powered by the Qwen 2.5-coder model. The AI Programmer offers features such as code completion and optimization, debugging assistance, code snippet search and batch unit test generation. It provides developers with an efficient and seamless coding experience, significantly enhancing productivity and creativity.
In addition to a broader range of models, a slew of new AI development tools is also accessible by global developers on Model Studio. These tools include Workflow, which breaks down complex tasks into subtasks to enhance workflow control, and Agent, which supports multi-agent collaboration for planning and execution tasks. Other tools such as RAG (Retrieval-Augmented Generation), which helps enhance the accuracy and reliability of generative AI models with external sources; Batch Reasoning, which generates responses simultaneously with multiple prompt inputs; AutoEval (Automated Model Evaluation), as well as model deployment and application observability services will be available by end of this month.
Upgraded Infrastructure for AI Development
In order to provide robust computing to facilitate AI and other critical workloads, Alibaba Cloud revealed that its 9th Generation Enterprise Elastic Compute Service (ECS) instance will be available in global markets starting April this year. The latest generation of ECS instances has notable performance enhancements compared to its previous iteration, including a 20% increase in computing efficiency. Additionally, by accelerating networks through eRDMA (elastic Remote Direct Memory Access), its performance in supporting high-performance computing, search recommendations, and Redis databases can be further improved by up to 50%.
The cloud pioneer has also announced its innovative Alibaba Cloud Container Compute Service (ACS) is now available for international customers starting from January 2025. Designed for simplified and optimized workload deployment using container technology, the ACS integrates container services with underlying cloud computing resources, significantly reducing costs and technical complexity and enabling developers to focus on innovation rather than infrastructure management.
New GenAI Program to Spark Creativity
To foster innovation, Alibaba Cloud introduced the Alibaba Cloud GenAI Empowerment Program, a dedicated support program for global developers and startups leveraging its Qwen models to build generative AI applications. Participants in the program can gain support including free cloud credits, training workshops, invitation to tech shows and demo days, as well as product co-marketing opportunities. This program is designed to help developers and startups accelerate their generative AI projects while connecting with a broader ecosystem of innovators.
Global Developers and Customers Leveraging Qwen for Cutting-Edge Applications
Axcxept, a Japanese company specializing in AI products such as voice assistants, has developed an open-source, lightweight AI model called EZO based on Qwen 2.5 LLM. The EZO model outperforms state-of-the-art (SOTA) models in areas such as coding, information extraction, math, reasoning, roleplay, and writing in Japanese. With low latency and robust performance, EZO is tailored to serve industries such as healthcare and public institutions in Japan, ensuring secure and efficient AI applications.
“Qwen 2.5 has significantly enhanced its ability to process Japanese, giving it a competitive edge over other models. With Axcxept’s proprietary training process, we have developed a Japanese LLM that delivers unmatched accuracy,” said Kazuya Hodatsu, CEO of Axcxept Inc.
OxValue.AI, a deep-tech venture from the University of Oxford, uses Alibaba Cloud’s Qwen-based multimodal AI models for AI-driven company valuation services. By processing and analyzing text and audio data related to financing, R&D, and operations, OxValue achieves precise and cost-efficient valuation assessments tailored to corporate clients.
“Processing diverse data sources is essential for our valuation services. With the support of Alibaba Cloud’s AI models, we’ve significantly improved the quality and efficiency of this process. By collaborating with Alibaba Cloud, we’re able to deliver greater value to our corporate clients,” said Professor Xiaolan Fu, Founder of OxValue.AI.
