Alibaba rolls out Qwen3.5 to take on OpenAI and Anthropic

Alibaba Qwen

• Alibaba launches Qwen3.5 AI with advanced multimodal features
• Qwen3.5 supports text, image, and video in one system
• Open-weight and hosted versions now support 201 languages

Did our AI summary help?

Alibaba has released its latest Qwen3.5 AI model series, introducing stronger multimodal and agent-focused capabilities as rival Chinese firms race to push out new models.

The Qwen3.5 lineup includes both an open-weight model and a hosted version. The open-weight release allows developers to download, fine-tune, and deploy the model on their own infrastructure, while the hosted variant runs on Alibaba’s cloud servers via its Model Studio platform.

Both versions were released on Monday, just ahead of the Chinese New Year, and arrive only a week after Alibaba launched a separate AI model aimed at robotics applications.

According to the company, Qwen3.5 delivers improvements in both performance and cost efficiency. It has been built with native multimodal capabilities, allowing it to process text, images, and video within a single system, rather than relying on separate models stitched together.

Alibaba is also leaning into one of the biggest AI trends of the year: agentic systems. Qwen3.5 supports advanced coding and agent capabilities and is compatible with open-source AI agents such as OpenClaw, which has recently gained traction among developers.

AI agents are designed to autonomously complete multi-step tasks with minimal human input, a development that has raised questions about how much traditional software-as-a-service tools may be disrupted. Interest in agents surged further after Anthropic released new agent tools earlier this year.

Alibaba’s move comes as Chinese competitors accelerate their own releases. ByteDance and Zhipu AI have both launched upgraded models in recent days, many of them focused on supporting more advanced agent behaviour.

The open-weight Qwen3.5 model reportedly includes 397 billion parameters, the internal variables that influence how AI systems learn and reason. While this is fewer than Alibaba’s previous flagship model, the company says internal benchmarks show significant gains in performance.

Alibaba has published benchmark results claiming Qwen3.5 performs on par with leading models from OpenAI, Anthropic, and Google DeepMind, though these comparisons are self-reported and could not be independently verified.

The hosted version, called Qwen3.5-Plus, is also claimed to match top-tier international models in performance. Alibaba said the new generation now supports 201 languages and dialects, a sharp increase from the previous model’s 82.

More open-weight releases are expected during the Chinese New Year period, according to Lin Junyang, technical lead of Alibaba Cloud’s Qwen team, who shared the update on social media.

The rapid pace of releases reflects growing global competition around AI agents. Following Anthropic’s recent announcements, OpenAI CEO Sam Altman confirmed that the creator of OpenClaw would be joining the company. Meanwhile, Google DeepMind chief Demis Hassabis said last month that Chinese AI models are only “months” behind their Western counterparts.