Alibaba Cloud

Alibaba Cloud (Aliyun) is the cloud computing arm of Alibaba Group and the largest cloud provider in Asia Pacific. Founded in 2009, it serves millions of customers across more than 200 countries and regions, offering a comprehensive suite of cloud infrastructure, data intelligence, and AI services.

Alibaba Cloud's AI portfolio centers on the Qwen (Tongyi Qianwen) family of large language models, licensed by tier. The Apache 2.0 line, led by Qwen 3.6, spans from sub-billion-parameter edge models upward and has made Alibaba one of the most important contributors to open AI. Above it sits Qwen3.8 Max, announced in August 2026 — a mixture-of-experts model with 2.4 trillion total parameters and roughly 95 billion active per request. Its weights were published on August 12, 2026 under a custom Qwen license rather than Apache 2.0, making it the second-largest open-weight release to date behind Moonshot's Kimi K3. The license gates scale rather than access: a model-as-a-service business whose revenue exceeds $50 million over twelve consecutive months must obtain a separate agreement from Qwen before commercial use. The smaller Qwen3.8-27B shipped on August 14, 2026 under plain Apache 2.0, with none of those thresholds. Size does not predict terms, though: Qwen3.8-Flash-Next — 125 billion total parameters with roughly 6 billion active per token, published on August 27, 2026 as a technology preview of the Qwen 4 architecture — carries the Qwen Community License 1.0, whose model-as-a-service clause requires a separate agreement from Alibaba at any size, with no revenue floor at all. Specialized variants cover coding (Qwen-Coder), math (Qwen-Math), and vision (Qwen-VL). The company also offers Model Studio for fine-tuning, PAI (Platform for AI) for ML operations, and a broad set of pre-built AI services for NLP, computer vision, and speech.

As the infrastructure backbone for Alibaba's e-commerce, logistics, and fintech operations, Alibaba Cloud brings massive-scale AI deployment experience. Its aggressive open-sourcing of Qwen models has made it one of the most important contributors to the open-weight AI ecosystem alongside Meta and Mistral.

Qwen's most consequential distribution win came in July 2026, when China's Cyberspace Administration cleared Apple Intelligence for release in China with Qwen as the model powering text and image understanding and generation across iOS, iPadOS, macOS, and visionOS for Chinese users. Apple evaluated Baidu, DeepSeek, and ByteDance before selecting Alibaba, with Baidu contributing a smaller model alongside Qwen. The arrangement puts Alibaba's models in front of one of the largest premium device populations in the world, and makes Alibaba Cloud the default AI layer for the iPhone in Apple's most important market outside the United States.

Alibaba published weights for two very different models two days apart in August 2026. Qwen3.8-Max, the 2.4-trillion-parameter flagship, landed on August 12 under a custom Qwen license requiring a separate agreement for any model-as-a-service business above 50 million dollars in revenue. Qwen3.8-27B followed on August 14 under a plain Apache 2.0 license with no thresholds at all — a multimodal 27-billion-parameter model with a 262,144-token context window that scores 61.7 on SWE-bench Pro against the flagship's 67.7. The pairing states Alibaba's licensing-by-tier strategy plainly: the largest model is downloadable but gated, while the compact one is genuinely unrestricted.

On September 10, 2026 Anthropic named Alibaba in a threat intelligence report as the source of the largest distillation campaign it has measured against Claude. Anthropic says operators affiliated with Alibaba injected a fixed prompt into each request to force Claude to write out its reasoning before answering, saved those traces, and converted them into supervised fine-tuning data used to train Qwen 3.5, 3.6 and 3.7 — more than 151 million exchanges between May and July 2026, peaking near three million a day, routed through pools of fraudulent accounts using residential proxies, disposable email addresses and virtual-card payments. It also says Alibaba used Claude to help build its own reinforcement learning environments and model architecture research. Alibaba had not publicly responded at publication, and Anthropic is a direct competitor, so this is an allegation supported by one party's telemetry rather than an established finding.