Alibaba on Monday unveiled its new Qwen3.5 AI model series, releasing both an open-weight version and a hosted version through its cloud platform, as competition in China’s AI market accelerates. The models were made available on the eve of the Chinese New Year, just one week after the company introduced a separate AI model designed for robots.
The open-weight version allows users to download, run, fine-tune and deploy the model on their own infrastructure. Alongside it, Alibaba launched a hosted option — called Qwen-3.5-Plus — via its Model Studio cloud platform, enabling customers to access the system directly on Alibaba’s servers.
Alibaba said Qwen3.5 delivers improvements in both performance and cost compared with prior versions. The company also emphasized that the model was built with “native multimodal capabilities,” meaning it can process text, images and video within a single system.
The release leans heavily into one of the industry’s fastest-moving areas: AI agents. Qwen3.5 supports new coding and agentic capabilities and is compatible with open-source AI agents such as those from OpenClaw, which have recently gained traction. AI agents are systems designed to independently take actions and complete multi-step tasks on a user’s behalf with minimal supervision.
Interest in such systems has intensified in recent weeks following the rollout of new agent tools by U.S.-based Anthropic. The idea that AI agents could potentially disrupt software-as-a-service providers and other internet business models has unsettled markets. Marc Einstein, research director at Counterpoint Research, told CNBC that AI companies are preparing for the possibility that agents could “upend traditional Internet business models,” adding that firms not ready for that shift could face severe consequences.
Alibaba’s domestic rivals are moving quickly as well. ByteDance and Zhipu AI both released upgraded models in the past week aimed at expanding agent capabilities, underscoring how crowded China’s AI field has become.
Technically, Alibaba said the open-weight Qwen3.5 model includes 397 billion parameters — the variables that shape how an AI system learns and reasons. While that figure is lower than its previous flagship model, the company reported significant performance gains based on its own benchmark evaluations. Alibaba shared test results indicating that Qwen3.5 performs on par with leading models from OpenAI, Anthropic and Google DeepMind, though those comparisons were self-reported and could not be independently verified by CNBC.
The hosted Qwen-3.5-Plus model, available through Alibaba Cloud’s Model Studio, was also described by the company as delivering performance comparable to top global competitors.
Another notable update is language coverage. The new Qwen3.5 models support 201 languages and dialects, up from 82 in the previous generation — a shift that Counterpoint’s Einstein said reflects Alibaba’s broader global ambitions.
Alibaba is expected to release additional open-weight models during the Chinese New Year period, according to Lin Junyang, technical lead of Alibaba Cloud’s Qwen team, who shared the update in a social media post.
The launch comes as global AI competition continues to tighten. Following Anthropic’s latest agent tools, American AI companies have accelerated development in similar areas. OpenAI CEO Sam Altman said Sunday that the creator of OpenClaw would be joining OpenAI. Meanwhile, Google DeepMind CEO Demis Hassabis said last month that Chinese AI models are only “months” behind Western rivals.
With Qwen3.5, Alibaba is positioning itself squarely in that race — emphasizing agent capabilities, broader language reach and multimodal functionality as the next phase of competition in China’s fast-moving AI market.
This analysis is based on reporting from CNBC.
Image courtesy of Qwen.
This article was generated with AI assistance and reviewed for accuracy and quality.
