Alibaba Cloud adds Qwen 3.5 and GLM-5 for coding

Alibaba Cloud is turning into a showcase for Chinese AI: the platform has integrated four leading open language models at once — Qwen 3.5, GLM-5, MiniMax M2.5 and Kimi K2.5. All four share one trait: they were built specifically for programming tasks, and their results come close to the best proprietary offerings on the market. For developers working on Alibaba's cloud infrastructure, this means access to a whole arsenal of tools without having to switch between several providers.

The context of this announcement matters. Over the past year and a half, Chinese tech companies have steadily released models that previously stayed in the shadow of their Western counterparts. Alibaba's own Qwen series, GLM from Tsinghua University and Zhipu AI, and the work of the MiniMax and Moonshot AI teams are not scattered experiments but a steadily built ecosystem.

Now Alibaba Cloud is consolidating these efforts under one roof, effectively staking a claim to the role of central hub for open-source AI not only in Asia but on the global market. The timing is no accident: Western enterprise users are increasingly looking for alternatives to expensive APIs from OpenAI and Anthropic, and regulatory uncertainty around closed models is pushing companies toward open solutions with predictable licensing.

Each of the four models fills its own niche in the new offering. Qwen 3.5 is the flagship of Alibaba's own lineup, strengthened for generating and debugging code in many programming languages.

GLM-5 from Zhipu AI focuses on complex reasoning and multi-step instructions, which is critical for automating software development. MiniMax M2.5 brings extended long-context capabilities to the mix — the ability to hold entire large codebases in memory without losing the coherence of its reasoning.

Kimi K2.5 from Moonshot AI covers agentic scenarios: the model can break a task into subtasks on its own, plan steps and carry them out in sequence, which makes it a promising foundation for autonomous coding agents. Together, the four systems cover nearly the full range of tasks a development team faces, from writing individual functions to comprehensive refactoring of large projects.

The implications of this move for the industry go beyond simply expanding a catalog. First, Alibaba Cloud creates direct competition for services such as Amazon Bedrock and Azure AI Studio, which aggregate models from different vendors under a single API. The difference is that the Chinese platform bets exclusively on open-source models, reducing customer dependence on a vendor and making it possible to deploy the same systems locally if needed.

Second, concentrating top Chinese models in one place makes them easier to test and compare for international corporations that are eyeing these developments but are not ready to build an integration with each provider separately. Third, the very fact of such consolidation signals the maturity of China's AI segment: the market has moved from a race of publications to commercial deployment.

For developers, the practical gain is already clear. Instead of juggling accounts at Zhipu AI, MiniMax and Moonshot, a single Alibaba Cloud account is enough, with unified billing, monitoring and scaling. This cuts operational load and lets teams focus on the product rather than on administering infrastructure.

The main question facing the market now is how durable this advantage will prove. The open-source landscape changes fast: models considered cutting-edge today risk landing in the middle of the benchmark tables in six months. Alibaba Cloud is betting not on specific checkpoints but on a platform effect — the more developers get used to working with its infrastructure, the harder it will be for competitors to pull them away. If the platform keeps integrating new models as promptly as they are released, it has every chance of securing the status of de facto standard for working with Chinese open-source systems across the global market.