On March 6, China’s AI agent Manus trended on Chinese social media platform Weibo. According to its team, Manus is an autonomous AI agent designed to handle complex tasks beyond traditional AI assistants by delivering complete task results. A four-minute demo showcased its capabilities, including screening resumes like a human intern. Manus achieved state-of-the-art (SOTA) results in GAIA benchmark tests. This may be the first AI agent product to signal the beginning of China’s AI agent era.
Why it matters: Manus’ rise reflects growing interest in AI agents capable of independently executing complex tasks, a shift from traditional AI assistants that primarily provide suggestions or answers.
Details:
- Manus, developed by a team led by entrepreneur Xiao Hong, has gained significant attention on Chinese social media. Unlike conventional AI assistants, it autonomously completes tasks rather than merely offering recommendations.
- The product achieved a state-of-the-art (SOTA) performance across all three difficulty levels of the GAIA benchmark, which evaluates AI assistants’ ability to solve real-world problems.
- Manus remains in beta testing, with users required to apply for access. Reports indicate high demand, with invite codes being resold on secondary markets for up to RMB 100,000 ($13,900).
- The company has denied any official paid channels for obtaining invite codes.
Context: China’s AI sector is experiencing intensified competition in AI agent development, with companies seeking to push beyond traditional chatbot functionalities. Manus’ emergence follows the rise of DeepSeek, another domestic AI model that recently gained significant traction. Globally, OpenAI’s announcement of a $20,000 monthly subscription for an enterprise-grade AI agent, signals growing commercial interest in autonomous AI.
- Xiao Hong, the founder behind Manus, previously launched AI assistant Monica, which integrates multiple large language models (LLMs) and initially gained popularity in overseas markets. A Chinese version of Monica, supporting deep reasoning and real-time search capabilities, entered beta testing in February.
0 Commentaires