Chinese AI Startups Are Gaining Ground Fast
Chinese startups are making serious moves in the AI space. Butterfly Effect and Zhipu have launched tools that claim to outperform OpenAI’s Deep Research on various benchmarks.
Both startups are racing alongside giants like Alibaba and ByteDance, who are pushing their own agents powered by proprietary models. The stakes are rising as global investments in AI technology surge. An IBM survey reported that over 60% of CEOs are ramping up their use of AI agents. 
Agent capabilities range from travel planning to app development. But not all are created equal. A U.S. market intelligence report warns that many Chinese companies are labeling their products as AI agents just for the hype, leading to inconsistent quality.
Experts point out the varying performance levels are largely linked to the foundational models behind these tools. Zhou Yu, a computer science professor at Columbia University, commented:
"LLMs are a sort of fundamental power. If the models are getting better, the usability of the agents will improve as well — a high tide lifts all boats."
Despite U.S. companies currently leading in AI model development, the competitive edge is narrowing. Sayash Kapoor, co-author of AI Snake Oil, noted:
"China lags significantly behind the U.S. in digitization of business processes, cloud computing adoption, and workforce training."
Here’s a quick rundown of key players in the race:
Manus – Butterfly Effect
Manus from Butterfly Effect launched its invite-only general-purpose AI agent in March. It aims to handle real-world tasks from itinerary planning to business analysis. Claims suggest it outperforms OpenAI’s Deep Research in certain metrics, but reviews are mixed. The company recently secured $75 million in funding.
Quark – Alibaba
Quark, Alibaba’s all-in-one AI agent, also launched in March. It tackles tasks like academic research, report writing, and image generation. The app skyrocketed to 149 million monthly active users in April, yet a media test found its output to be too simplistic for professional use.
AutoGLM Rumination – Zhipu
Zhipu’s AutoGLM Rumination debuted in March, showing promise in tasks like price comparisons and report generation. However, performance tests indicated it struggles with filtering misinformation.
Coze – ByteDance
Coze, launched by ByteDance in April, focuses on data analysis, report writing, and app creation. In side-by-side tests, Coze achieved 81% task completion versus Manus’ 92%, showing fierce competition ahead.
Chinese tech companies are stepping up their game in AI, and the global race is tightening. Expect more updates as they strive for dominance in this evolving landscape.
