Recasts with details the release of AI agent Manus

By Che Pan and Brenda Goh

BEIJING, March 6 (Reuters) - Alibaba Group's 9988.HK release of an artificial intelligence $(AI)$ reasoning model, which it said was on par with global hit DeepSeek's R1, drove its Hong Kong-listed shares more than 8% higher on Thursday.

In a post on X, the e-commerce leader's AI unit said its QwQ-32B, with 32 billion parameters, can achieve performance comparable to DeepSeek's R1 model that has 671 billion parameters.

As the world races to adopt AI models, the Chinese government on Wednesday pledged increased support for industries including artificial intelligence, humanoid robots and 6G telecom.

Alibaba said its new model is accessible via its chatbot service, Qwen Chat, for which users can choose various Qwen models including Qwen2.5-Max, the most powerful language model in the Qwen series.

The QwQ-32B demonstrated capabilities in mathematical reasoning, coding and general problem-solving in benchmark tests, performing close to top models such as OpenAI's o1 mini and DeepSeek's R1, Alibaba said further.

DeepSeek has emerged as the poster child of China's AI prowess, rivaling top models from OpenAI for a small fraction of their cost with less powerful computing.

Another AI release that attracted significant attention on Thursday was the release of an AI agent called Manus by Chinese startup Monica, which said it beat ChatGPT maker OpenAI's Deep Research on a benchmark for AI assistants.

An AI agent is a more advanced version of a chatbot and according to use cases listed on its website, Manus can help users to make a travel plan to Japan, or conduct a comparative analysis of insurance policies.

The AI agent is for now by invitation only. A video released by the Chinese startup on X late on Wednesday demonstrating how it worked received over 280,000 views by Thursday, with many users asking for invites.

(Reporting by Che Pan and Brenda Goh; Editing by Mrigank Dhaniwala and Barbara Lewis)