Alibaba Group Holding Ltd. has unveiled Qwen3.8-Max, its most advanced artificial intelligence model to date, ahead of a planned open-weight release to the public scheduled for next week.

The model features 2.4 trillion parameters and outperforms Moonshot's recently launched Kimi K3 across multiple benchmarks. The launch also signals Alibaba's return to its open-source origins, following a period in which several of its top-tier AI models were released as proprietary offerings earlier this year.

Qwen3.8-Max is currently available through Alibaba Cloud's Model Studio APIs and via QwenWork, the company's AI assistant platform for enterprise customers. The open-weight release planned for next week will allow users to download the model weights and fine-tune the system themselves. The move places Alibaba in direct competition with U.S. rivals OpenAI and Anthropic.

Unlike proprietary AI systems that offer API access exclusively, an open-weight release enables developers and businesses to download model weights, customize them for proprietary AI systems, and build applications on top of them. This approach gives enterprises greater control over deployment, privacy, and performance optimization, while allowing researchers to replicate and extend Alibaba's work without starting from scratch. The strategy aligns with Alibaba's broader objective of accelerating enterprise adoption across its AI ecosystem. Open-weight releases have become a key competitive lever in the AI industry—Meta's Llama series demonstrated that freely available model weights can attract large developer communities and drive adoption of the releasing company's cloud and tooling platforms.

Benchmark Performance and Autonomous Capabilities

According to Alibaba, Qwen3.8-Max has achieved results comparable to—or in some cases exceeding—those of Anthropic's Fable 5 across several benchmarks. The model can transform complex data sources, including video streams and lengthy technical manuals, into searchable archives.

Additional capabilities include recreating applications from screenshots, designing interactive games and educational content, and converting flat floor plans into immersive 3D visualizations. Alibaba stated that the model excels at autonomous coding, complex research, and other tasks requiring sustained processing over extended periods. To demonstrate its autonomy, the model independently managed an in-house software project for 16 consecutive days. The system also improves efficiency through selective activation of only the necessary components—a Mixture-of-Experts design that allows the model to scale to trillions of parameters while keeping inference costs and latency manageable, an approach also used in frontier models from other leading labs.

In user-voted evaluation arenas, Alibaba's model has secured top rankings, trailing only a select group of premium Claude models. It placed fifth in text tasks and second in vision tasks. The model was initially released on July 19 through Alibaba's subscription plan and its Qoder ecosystem, with the company asserting at the time that its performance ranked second only to Fable 5.

Following the debut announcement, Alibaba's shares rose by as much as 7.3%, marking their strongest performance in nearly a month. The market reaction underscores investors' growing view of AI as a primary long-term growth driver for the company. Alibaba has committed substantial resources to cloud computing and generative AI, with expectations that improved model performance will attract more enterprise customers to its cloud platform and developer ecosystem.

Escalating China–U.S. AI Competition

The launch reflects the intensifying rivalry between Chinese and American AI developers, both racing to release increasingly capable foundation models at an accelerated pace. Companies are now differentiating themselves not only through benchmark scores but also through coding proficiency, multimodal functionality, enterprise integration, and the availability of open-source models that developers can freely adapt.

The Qwen3.8-Max debut adds further momentum to China's rapid advancement in the sector, arriving shortly after Moonshot's Kimi model drew significant attention across Western technology centers. Chinese laboratories continue working to close the gap with OpenAI and Anthropic while operating with comparatively fewer computing resources, a constraint driven in part by U.S. export controls on advanced AI chips that restrict access to the latest NVIDIA GPUs and other high-performance silicon. In recent developments, DeepSeek expanded access to its V4 Flash model, while ByteDance and MiniMax both launched new video generation models on Friday.

Vey-Sern Ling, managing director at Union Bancaire Privée, argued that Chinese models are approaching parity with leading U.S. systems. He stated: "Many investors continue to underestimate Chinese AI models because of US chip restrictions or general skepticism. In reality, the gap is probably much closer and narrowing fast. Alibaba's Qwen 3.8 is another proof point, following Kimi K3."

The technological rivalry between the two nations has intensified in recent months. The U.S. government previously blacklisted foreign access to Anthropic's Fable 5 and Mythos 5 models, temporarily restricting them as a precautionary measure to prevent potential security circumvention by Chinese authorities. Anthropic subsequently restored access to its overseas user base on Washington's orders, while noting that it was implementing stronger safeguards and had made commitments to the government regarding technology protection.

More recently, Anthropic accused Chinese developers of conducting distillation attacks. In a letter to two U.S. lawmakers, the San Francisco-based company stated that operators linked to Alibaba had used thousands of fake accounts to conduct nearly 29 million exchanges with Claude—describing it as the largest extraction campaign it has encountered. Distillation attacks involve systematically querying a model to extract its outputs and using them to train a competing model, effectively replicating the target system's capabilities without authorized access. Anthropic urged lawmakers to hold the entities behind these attacks accountable and to strengthen measures preventing the unauthorized appropriation of U.S. technology.

According to Anthropic, the operation was highly targeted, focusing specifically on Claude's capabilities in managing complex projects and high-level reasoning. The company stated: "Distillation attacks turn hundreds of billions of dollars in American investment and [research and development] into a massive subsidy for our geopolitical competitors."