Alibaba Cloud updates Coding Plan: new Qianwen and GLM models

Alibaba Cloud has expanded its Coding Plan platform by adding support for several next-generation language models — the flagship Qianwen 3.5 and GLM-4.7 from Zhipu AI. The update looks technical, but behind it lies a much larger bet: Chinese technology giants are seriously aiming to displace GitHub Copilot and similar Western tools in domestic and regional markets.

Coding Plan is Alibaba's cloud environment oriented toward developers, combining tools for collaborative code work, CI/CD pipelines, and built-in AI assistance. Until recently, the choice of language models within the platform was limited. The new update radically changes this picture: users can now switch between several models depending on the task, whether it's generating code from scratch, auto-completion, or finding errors in already written solutions.

Qianwen 3.5 is Alibaba's own flagship development. The model was created with a focus on working with code in dozens of programming languages and can not only complete lines but also understand the context of the entire project. According to the company, it demonstrates improved understanding of long contexts — a critical property for working with large codebases where you need to keep the architecture of several modules in mind simultaneously. GLM-4.7, developed by startup Zhipu AI with support from Tsinghua University, takes a somewhat different path: the model emphasizes reasoning accuracy and the ability to explain code logic in natural language, which is especially useful in team projects and code reviews.

The inclusion of GLM-4.7 in Alibaba's ecosystem is notable in itself. This is not just a technical expansion — it is evidence that major platforms in China are beginning to build partnership models instead of closing themselves off exclusively to their own developments. Zhipu AI has long been considered one of the country's most serious AI startups, and the integration of its model into Coding Plan gives the young company access to a multi-million strong audience of developers already working in Alibaba's cloud. For the platform, this is also beneficial: model diversity means that each user can find a tool that fits organically into their workflow.

The competitive context is equally important. Baidu with its Comate platform and Huawei with CodeArts have long claimed to be the leading AI assistants for Chinese developers. This market is growing rapidly: according to analysts, the adoption of AI tools in development workflows in the Asia-Pacific region accelerated after major companies recorded actual reductions in time spent on routine code writing — from 20 to 40 percent depending on the task. Alibaba clearly does not want to cede this market to competitors and is betting on model diversity as a competitive advantage: instead of a single solution — an ecosystem where the user chooses.

What this means for developers outside China remains an open question. Alibaba Cloud is actively expanding its presence in Southeast Asia, the Middle East, and Europe, and such updates sooner or later become available to international audiences. Qianwen in various versions has already demonstrated competitive results in open benchmarks for code generation, making it potentially interesting not only for the Chinese market. At the same time, GLM remains less familiar to the international community, although the academic roots of the model are a serious argument in favor of its technical credibility.

The Coding Plan update is not an isolated event, but another step in a large-scale restructuring of how software code is written in the era of large language models. Tools like GitHub Copilot have changed developer expectations: an AI assistant today is perceived not as an addition but as part of the standard working environment. Alibaba understands this and responds not simply by updating one model, but by creating a multi-model environment capable of adapting to different requirements. How quickly the platform will be able to gain critical mass of users and how much the actual performance of the models will match marketing promises — the coming months will tell.