By Jessica Williams
BEIJING — Alibaba Group unveiled its latest artificial intelligence model series, Qwen3.5, on Monday, just ahead of the Chinese New Year celebrations, intensifying the race among Chinese tech giants to advance AI agents capable of autonomous task execution. The release comes amid a flurry of model upgrades in China over the past week, as companies pivot from traditional chatbots to more sophisticated systems that can handle multi-step actions with minimal human oversight.
According to Alibaba, the Qwen3.5 series includes an open-weight version, allowing developers worldwide to download, fine-tune, and deploy the model on their own infrastructure. A hosted version, dubbed Qwen-3.5-Plus, is available through Alibaba's cloud platform, Model Studio, enabling users to access it via the company's servers without needing local hardware. Both variants were launched on February 16, 2026, the eve of the Lunar New Year, following Alibaba's introduction of a robot-focused AI model just seven days earlier.
The new models boast native multimodal capabilities, meaning they can process and understand text, images, and video within a single unified system. Alibaba emphasized improvements in performance and cost efficiency, with the open-weight model featuring 397 billion parameters — the adjustable variables that influence how an AI learns and reasons. Although this parameter count is lower than the previous flagship's, Alibaba reported significant gains based on internal benchmark evaluations.
In self-conducted tests, Alibaba claimed that Qwen3.5 performs on par with leading models from OpenAI, Anthropic, and Google DeepMind. The hosted Qwen-3.5-Plus also purportedly matches these competitors, according to the company. However, CNBC was unable to independently verify these benchmarks, highlighting the challenges in objectively assessing AI performance amid rapid industry advancements.
Supporting 201 languages and dialects — a substantial increase from the prior generation's 82 — Qwen3.5 is designed for global accessibility. Alibaba highlighted its compatibility with open-source AI agents, including those from OpenClaw, a framework that has gained traction recently for enabling autonomous AI systems. AI agents, as defined by industry experts, are software entities that can independently execute actions and complete complex, multi-step tasks on behalf of users, often requiring little to no supervision.
This focus on agentic capabilities aligns with a broader trend reshaping the AI landscape in 2026. Just weeks ago, American AI firm Anthropic introduced new tools for its Claude model that enhance agent functionalities, sparking widespread interest. The potential of these agents to automate workflows has unsettled markets, with some analysts suggesting they could disrupt software-as-a-service providers and other sectors reliant on human-mediated processes.
Alibaba's move is part of a competitive surge in China. ByteDance, the parent of TikTok, and Zhipu AI both rolled out upgraded models in the past week, each emphasizing enhanced support for agent-based operations. This wave of releases underscores Beijing's push to catch up in AI innovation, where domestic firms are vying not only with each other but also with U.S. counterparts.
Lin Junyang, technical lead of Alibaba Cloud's Qwen team, announced via social media that more open-weight models are slated for release during the Chinese New Year period. "Alibaba is expected to release more open-weight models during this Chinese New Year," Junyang wrote in a post shared on February 17, 2026, signaling the company's aggressive development pace.
The timing of Alibaba's announcement coincides with heightened activity among global AI leaders. On Sunday, OpenAI CEO Sam Altman revealed that the creator of OpenClaw would be joining the company, bolstering OpenAI's agent development efforts in the wake of Anthropic's breakthroughs. This talent acquisition reflects the fierce competition for expertise in building next-generation AI systems.
Contextually, the shift toward AI agents builds on years of investment in foundational models. Alibaba's Qwen series, first introduced in 2023, has evolved rapidly, with each iteration incorporating feedback from developers and users. The addition of coding enhancements in Qwen3.5 allows the model to assist in programming tasks more effectively, further enabling agentic applications like automated software debugging or content generation pipelines.
From a geopolitical perspective, the advancements in Chinese AI have drawn scrutiny from Western observers. Last month, Demis Hassabis, head of Google DeepMind, told CNBC that Chinese models lag just "months" behind their Western rivals. Hassabis's assessment, made in January 2026, suggests a narrowing gap, driven by China's vast data resources and state-backed research initiatives.
Yet, challenges remain. While Alibaba touts Qwen3.5's multimodal prowess, real-world deployment of AI agents raises questions about reliability, ethical safeguards, and integration with existing infrastructure. Industry watchers note that agent systems must navigate complex environments, from enterprise software to consumer apps, without introducing errors or biases that could amplify risks.
In the broader market, Alibaba's release could bolster its position in the cloud computing arena, where AI services are a key revenue driver. The company's Model Studio platform, hosting the Qwen-3.5-Plus, positions Alibaba to attract enterprise clients seeking scalable AI solutions. Competitors like Tencent and Baidu have similarly invested heavily, creating a vibrant ecosystem in Hangzhou and Shenzhen, Alibaba's home bases.
Looking ahead, the emphasis on open-weight models democratizes access, potentially fostering innovation among startups and researchers outside major tech hubs. However, it also invites concerns over intellectual property and security, as freely available models could be adapted for unintended uses. Alibaba has not detailed specific safeguards in its Qwen3.5 documentation, though the company maintains compliance with international AI ethics standards.
As the Chinese New Year unfolds, with festivities marking the Year of the Dragon starting February 17, 2026, Alibaba's announcements serve as a symbolic launchpad for its AI ambitions. With rivals accelerating agent development, the coming months will likely see further iterations, testing the limits of what autonomous AI can achieve. For now, Qwen3.5 stands as a testament to China's evolving role in the global AI race, blending open innovation with strategic competition.