Chinese companies are gaining ground in the global race to develop the next generation of artificial intelligence — not just chatbots, but autonomous agents designed to handle complex tasks with minimal human input.

In recent months, Chinese tech giants and startups have launched their own agents. Startups Butterfly Effect and Zhipu claim their tools outperform OpenAI’s Deep Research in some metrics. Giants such as Alibaba and ByteDanceiByteDanceByteDance is a Chinese internet technology company that owns TikTok and Douyin, a Chinese version of TikTok with a successful e-commerce arm.READ MORE are powering agents with their in-house foundation AI models, going toe-to-toe with platforms from Microsoft, Google, Amazon, and others. 

Global investments in the technology are surging as businesses increasingly pursue automation to boost efficiency and cut costs. An IBM global survey found more than 60% of CEOs are already deploying and planning to scale their use of AI agents. 

Agent capabilities vary widely — from travel planning to app development — and a lack of common benchmarks also makes it difficult to evaluate and compare performance. 

“Many Chinese tech companies are labeling their products as AI agents just to capitalize on the hype … leading to uneven product quality,” a report by U.S. market intelligence firm IDC said. Agent performance depends heavily on the foundation AI models they are built on, experts told Rest of World.

“LLMs are a sort of fundamental power,” Zhou Yu, associate professor of computer science at Columbia University and founder of an AI agent startup, told Rest of World, referring to the large language models that agents are built on. “If the models are getting better, the usability of the agents will improve as well — a high tide lifts all boats.” 

While U.S. companies still lead in AI model development, the gap is narrowing, according to Sayash Kapoor, co-author of AI Snake Oil, a book that examines claims about AI capabilities. But the “real competition” lies in effectively deploying AI agents across the economy, he said. 

“China lags significantly behind the U.S. in … digitization of business processes, cloud computing adoption, and workforce training,” Kapoor told Rest of World. 

Though still far from widespread use, these four companies are among the first in China to launch AI agents. 

Manus – Butterfly Effect

Manus, developed by Chinese startup Butterfly Effect, drew global attention after launching an invite-only general-purpose AI agent in March with English-language videos and other promotional material. 

The agent promises to execute a variety of real-world tasks — from itinerary planning to business analysis. Built using foundation models like Anthropic’s Claude and Alibaba’s Qwen, it claims to outperform OpenAI’s Deep Research on some benchmarks.

Despite the buzz, reviews have been mixed on whether Manus delivers on its claims. Still, the company reportedly raised $75 million in funding led by a Silicon Valley venture firm. Manus recently introduced paid subscription plans for public access. 

Quark – Alibaba

Quark, launched in March by Chinese tech giant Alibaba, bills itself as an “all-in-one” AI agent, powered by Alibaba’s own Qwen model.

It aims to fulfill a wide range of work-related tasks including academic research, medical diagnostics, image generation, presentation slides, and report writing.

The app quickly rose to the top of China’s AI app rankings in April, with 149 million monthly active users, even surpassing DeepSeek. A user test by Chinese tech media outlet 36Kr, however, found that while Quark could generate slides in just 10 seconds, the content was too simplistic to meet professional standards.

AutoGLM Rumination – Zhipu 

Zhipu, one of China’s top AI startups, launched its agent AutoGLM Rumination in March. 

In its introduction videos, the agent performed tasks such as comparing food delivery prices across apps and generating research reports through web searches. 

Backed heavily by government funding, the company has initiated the process of an initial public offering. Tests by a China-based tech consulting firm found the agent helpful in executing online searches, but not effective in screening results for false information.  

Coze – ByteDance 

ByteDance, TikTok’s parent company, launched AI agent Coze in April.

Coze offers four major functions: data analysis, report writing, and website and app creation. It has a separate finance-focused agent developed in partnership with Chinese investment platform Huatai Securities to analyze stock market data. 

Built primarily with ByteDance’s in-house Doubao 1.5 Pro model, Coze is designed to integrate other ByteDance platforms, such as TikTok’s Chinese sister app Douyin, in order to retain users. 

In a side-by-side test, Coze completed 81% of complex tasks successfully, compared to Manus’ 92%.