Alibaba Launches Qwen3.5 Amidst Shifting AI Agent Focus in China’s Chatbot Race

Alibaba Boosts AI Capabilities with Qwen3.5 Amidst Fierce Domestic Competition

Alibaba Group has unveiled its latest suite of large language models, Qwen3.5, marking a significant advancement in its artificial intelligence offerings. This new series incorporates enhanced functionalities, including “advanced, dynamic reasoning,” as the tech giant navigates an increasingly competitive AI landscape in China, where numerous models have been launched in rapid succession.

The Qwen3.5 series features an open-weight version, empowering developers and organizations to download, operate, fine-tune, and deploy the model on their proprietary infrastructure. Complementing this, Alibaba also offers a “hosted version” accessible via its cloud platform. These releases, timed just before the Lunar New Year, follow closely on the heels of Alibaba’s introduction of an AI model specifically engineered for robotics.

Alibaba emphasizes that Qwen3.5 delivers notable improvements in both performance and cost-efficiency. A key highlight is its “native multimodal capabilities,” enabling the model to process and understand text, images, and video concurrently within a single system. Embracing a prominent trend in the current AI development cycle, Qwen3.5 also supports advanced coding and agentic functionalities, demonstrating compatibility with popular open-source AI agent frameworks such as OpenClaw.

AI agents, designed to autonomously execute multi-step tasks on behalf of users with minimal human intervention, have recently captured significant market attention. This surge in interest was amplified by the release of new agent tools from American AI firm Anthropic, sparking discussions about their potential to disrupt various sectors, including Software as a Service (SaaS). This competitive pressure is evident as local rivals like ByteDance and Zhipu AI have also launched upgraded models emphasizing enhanced agent capabilities.

“AI companies are bracing for the possibility that AI agents could upend traditional internet business models,” noted Marc Einstein, Research Director at Counterpoint Research. “If this happens, the consequences for those who are not prepared will be severe, and Chinese AI companies are acutely aware of this.”

Alibaba’s open-weight Qwen3.5 model boasts 397 billion parameters. While this represents a reduction from its previous flagship model, the company asserts that benchmarks indicate significant performance enhancements. Alibaba has presented internal benchmark data suggesting that Qwen3.5’s performance rivals leading models from industry players like OpenAI, Anthropic, and Google DeepMind, though these comparisons are self-reported.

The company has also introduced a hosted model, Qwen-3.5-Plus, through its cloud-based Model Studio. Alibaba claims this version achieves performance parity with major competitors, a claim that CNBC has not independently verified.

Expanding its global reach, the new Qwen3.5 models demonstrate support for 201 languages and dialects, a substantial increase from the 82 supported by the previous generation. Einstein suggests this feature underscores Alibaba’s international ambitions. Lin Junyang, technical lead of Alibaba Cloud’s Qwen team, indicated via social media that further open-weight model releases are anticipated during the Chinese New Year period.

The acceleration in agentic capability development is a global phenomenon. Following Anthropic’s recent announcements, major American AI players are also intensifying their efforts. OpenAI CEO Sam Altman recently confirmed that the creator of OpenClaw would be joining the company. This follows Google DeepMind head Demis Hassabis’s earlier assessment that Chinese AI models were trailing Western rivals by only “months.”