Alibaba Launches Qwen-3.5 AI Models And Open-Weight Ahead Of Lunar New Year, As Competition Intensifies In China

Alibaba expands AI portfolio with multimodal Qwen-3.5 models. Image Credit: Getty Images
Share it:

Alibaba Group has published its latest series of AI models with improved features, as it experiences steep competition in the Chinese AI market with other models being rolled out over the last week.

The Qwen3.5 AI model is available in an open-weight variant, allowing users to download, run, fine-tune, and deploy it to their infrastructure. Alibaba had released a “hosted version,” referred to as the model, which can run on Alibaba’s own servers.

Both the models were launched on Monday, the eve of the Chinese New Year, and are only a week after Alibaba unveiled a new AI model aimed at robots.

The company pointed out that Qwen3.5 provides performance and cost benefits and was designed with “native multimodal capabilities” so that the models can interpret text, images, and video at the same time in a single system.

Bending towards one of the biggest AI trends of the year, the model also allows new coding and agentic functionalities, and can be used with open-source AI agents like those of OpenClaw, which have recently become popular.

AI agents refer to systems capable of making decisions and performing multi-step tasks autonomously at the behest of the user with minimum supervision.

These agents and their capabilities have become the focus of a lot of attention over the past few weeks, with an American AI company called Anthropic launching new agent tools. The prospects of the existence of these agents to oust the efforts of software as a service firms, among others, have shocked markets.

Local competitors of Alibaba, including ByteDance and Zhipu AI, also launched upgraded models over the last week with the purpose of facilitating additional agent abilities.

The company claimed that its new Qwen3.5 open-weight model has 397 billion parameters – variables that determine the way an AI system learns and reasons. Though it was lower than its past flagship model, the company claimed that the latest one was much better, according to self-reported benchmark assessments.

Alibaba ran benchmark tests indicating that Qwen-3.5 was as good at performance as the top models of OpenAI, Anthropic, and Google DeepMind, and that the comparisons were self-reported.

However, it also released a “hosted model” called the Qwen-3.5-Plus through its cloud platform Model Studio. Alibaba indicated that this version was also able to perform equally to leading competitors. CNBC could not independently verify those claims.

The new Qwen3.5 models also support 201 languages and dialects, up from the previous generation’s 82. Alibaba is projected to release more open-weight models during this Chinese New Year, Lin Junyang, technical lead of Alibaba Cloud’s Qwen team, said in a social media post.

Since the launch of the most recent Claude AI agent tools by Anthropic, other American AI giants have been working expeditiously on the creation of agentic capabilities. OpenAI CEO Sam Altman stated on Sunday that the creator of the OpenClaw would be joining the company.

Google DeepMind leader Demis Hassabis said last month in an interview with CNBC that Chinese AI models were only months behind their Western competitors.