
华为云码道CodeArts代码智能体上线鸿蒙编码大模型
华为云在智果 AgentArts 企业级智能体平台上宣布,旗下 AI 编码智能体「码道 CodeArts」面向鸿蒙开发场景上线全新鸿蒙编码大模型,目标是打造「最懂鸿蒙开发的 AI 编码智能体」。
据华为云披露的数据,新模型通过深度增强 ArkTS 与鸿蒙 API 生态能力,可使千行代码错误数最多降低约 80%,Token 消耗降低 20% 以上;同时它也是业内首个支持端侧模拟器直接调用的鸿蒙编码智能体,开发者可在本地模拟器中即写、即测、即看效果。
模型体系方面,码道 CodeArts 已搭载 GLM-4.7、GLM-5、GLM-5.1、GLM-5.2 等旗舰代码大模型,并在此基础上增训出 GLM-4.7-ArkTS-SPARK、GLM-5.2-ArkTS-SPARK 等鸿蒙专属变体,使模型在 ArkTS 语法、鸿蒙 API 调用模式与常见工程实践上实现深度适配。华为云更新日志显示,内置模型已由 GLM-4.7-ArkTS-SPARK 升级至 GLM-5.2-ArkTS-SPARK,长上下文理解与工程级编码能力明显增强。
功能上,码道 CodeArts 覆盖代码生成、项目级代码续写、研发知识问答、单元测试用例自动生成、规范驱动开发、Codebase 代码库索引以及专家技能 Skills 调用等能力,既面向个人开发者的轻量编码场景,也支持企业团队的整仓代码管理与全流程研发协同。
针对鸿蒙场景,码道 CodeArts 原生支持 ArkTS 这一鸿蒙官方推荐开发语言,并依托 Codebase 代码库索引对鸿蒙 SDK、组件库与工程实践做专属适配,据称可将代码调测错误数降低 50% 以上。
对鸿蒙开发者而言,码道 CodeArts 已不只是代码补全工具,而是贯穿需求分析、架构设计、编码实现、测试验证、规范检查乃至上架发布的智能助手:开发者可用自然语言描述需求,由智能体自动生成工程骨架与关键业务代码,再通过单元测试自动生成与规范驱动开发做质量控制,形成「从想法到上架」的闭环。
行业层面,华为云强调码道 CodeArts 面向金融、制造等对合规要求较高的行业提供全流程研发工具链、代码安全管控与合规保障,既可在本地或私有环境部署智能体,也可借助企业级权限管理与审计机制满足监管要求。随着公测面向个人开发者开放免费体验,中小团队与独立开发者可在不显著增加成本的前提下使用鸿蒙编码大模型进行原生应用开发。
本文来源:https://www.huxiu.com/ainews/15188.html
若有版权问题,请来信告知:13951431913#139.com 或在右方提交评论。
Huawei Cloud CodeArts Code Intelligence Agent launches Hongmeng coding large model
Huawei Cloud announced on the AgentArts enterprise intelligent agent platform of Zhiguo that its AI coding intelligent agent "CodeArts CodeDao" has launched a new Hongmeng coding large model for Hongmeng development scenarios, aiming to build "the most understanding of Hongmeng development AI coding intelligent agent".
According to data disclosed by Huawei Cloud, the new model, by deeply enhancing ArkTS and Hongmeng API ecosystem capabilities, can reduce the number of errors per thousand lines of code by up to about 80% and reduce token consumption by more than 20%; it is also the first industry-wide Hongmeng coding intelligent agent that supports direct invocation of on-device simulators, allowing developers to write, test, and see results instantly in a local simulator.
In terms of model system, CodeArts CodeDao has been equipped with flagship code large models such as GLM-4.7, GLM-5, GLM-5.1, GLM-5.2, and based on these, further trained to produce Hongmeng-specific variants such as GLM-4.7-ArkTS-SPARK and GLM-5.2-ArkTS-SPARK, enabling deep adaptation of the model to ArkTS syntax, Hongmeng API calling patterns, and common engineering practices. Huawei Cloud's update log shows that the built-in model has been upgraded from GLM-4.7-ArkTS-SPARK to GLM-5.2-ArkTS-SPARK, significantly enhancing long-context understanding and engineering-level coding capability.
Functionally, CodeArts CodeDao covers capabilities such as code generation, project-level code continuation, R&D knowledge Q&A, automatic generation of unit test cases, specification-driven development, Codebase code library indexing, and invocation of expert skills Skills, serving both lightweight coding scenarios for individual developers and supporting enterprise teams in full-repository code management and end-to-end R&D collaboration.
For Hongmeng scenarios, CodeArts CodeDao natively supports ArkTS, the Hongmeng officially recommended development language, and leverages the Codebase code library index to make dedicated adaptations to the Hongmeng SDK, component library, and engineering practices, allegedly reducing code debugging errors by more than 50%.
For Hongmeng developers, CodeArts CodeDao is no longer just a code completion tool; it is an intelligent assistant that runs through requirement analysis, architectural design, coding implementation, testing verification, specification checking, and even release publishing: developers can describe requirements in natural language, the agent automatically generates project scaffolding and key business code, then through auto-generated unit tests and specification-driven development performs quality control, forming a closed loop from 'idea to launch'.
At the industry level, Huawei Cloud emphasizes that CodeArts CodeDao provides a full-process R&D toolchain, code safety control, and compliance assurance for high-compliance industries such as finance and manufacturing; it can be deployed locally or in private environments, or leveraged with enterprise-level permission management and audit mechanisms to meet regulatory requirements. With the public beta opening free trial for individual developers, small and medium teams and independent developers can use the Hongmeng coding large model for native application development without significantly increasing costs.