China's AI scene is heating up, with new models rivaling top US players and rattling the stock market.
One of the latest contenders is Manus, an AI agent that some are calling China's next "DeepSeek moment."
But it's not alone. In March, Alibaba and Baidu unveiled AI models they said matched or outperformed DeepSeek while being more efficient.
With Chinese AI models gaining ground fast, here are five key players to watch.
1. DeepSeek's R1
DeepSeek was the first Chinese AI startup to shake up the industry. In January, its new model sent some US tech stocks plunging and raised questions about whether the US lead in AI has slipped.
The startup, based in Hang Zhou, was founded in 2023 by hedge fund entrepreneur Liang Wenfeng.
After buying thousands of Nvidia chips, Wenfeng built DeepSeek's R1 reasoning model on its V3 base model, training it on around 2,000 Nvidia H800 chips at an overall cost of roughly $5.6 million, the startup said.
That's a fraction of what top players like Meta and OpenAI are spending. Mark Zuckerberg has said Meta plans to spend over $60 billion this year as it doubles down on AI.
Despite its lower cost, its R1 model rivals top competitors like ChatGPT's o1, DeepSeek said.
Its impact on the AI chip market was immediate: Nvidia's stock plunged nearly 17%, wiping hundreds of billions from its market cap.
Microsoft CEO's, Satya Nadella, wrote on X in January: "Jevons paradox strikes again!"— suggesting that as AI becomes more efficient and accessible, demand will only skyrocket.
Marc Andreessen, the cofounder of storied Silicon Valley venture capital firm Andreessen Horowitz, took it a step further. He called DeepSeek's R1 "AI's Sputnik moment," likening it to the Soviet satellite launch that triggered the space race.
But DeepSeek's R1 has its limits. While it appears capable of doing much of what ChatGPT can, it couches answers related to sensitive issues in China.
2. Alibaba's QwQ-32B
Alibaba has become a key player to watch after it open-sourced its QwQ-32B model in early March. The company said the model used less data than DeepSeek.
The move sent Alibaba's stock surging 8% over two days, while Nvidia took another hit.
Alibaba said its QwQ-32B model, with one-fifth of the parameters of DeepSeek-R1, is designed for efficiency.
The model is now open-source on platforms, including Hugging Face. Open-source models allow for the free and open sharing of software to anyone for any purpose.
Alibaba's shares in Hong Kong and its secondary listing in New York are up nearly 67% so far this year, though its $353.9 billion market cap is still much smaller than Amazon's $2.08 trillion.
The company has been on a hot streak. Last month, its stock surged after a partnership with Apple to bring AI features to iPhones in China. Shares also climbed on news of Xi Jinping meeting with the country's tech moguls, including from Alibaba.
Early this month, the company announced plans to invest at least 380 billion Chinese yuan, or $53 billion, in cloud computing and AI infrastructure over the next three years.
3. Tencent's Yuanbao
Tencent's AI chatbot Yuanbao topped China's iOS App Store in early March, surpassing DeepSeek as the most downloaded free app, Bloomberg reported.
Tencent, based in Shenzhen, operates China's biggest social media app WeChat — which is used by nearly 1.4 billion people.
Last month, the company added a download button for Yuanbao in WeChat, making the AI chatbot more accessible to the app's users, the South China Morning Post reported.
Yuanbao is now the third most downloaded free app in China's iOS App Store, leading behind DeepSeek and ByteDance's Doubao. Doubao is a ChatGPT-like conversational bot.
In February, Tencent integrated DeepSeek's R1 into several of its products, including WeChat's search feature and a virtual companion in its evergreen game Peacekeepr Elite, Bloomberg reported last week.
The report added that Tencent also launched a Hunyuan Turbo edition that it says delivers answers faster than DeepSeek.
4. Manus
Manus has just become China's latest AI sensation.
According to multiple reports, it was built by Chinese startup Monica, which appears to be a subsidiary of Butterfly Effect. Manus' privacy policy said that Butterfly Effect is a Singapore-registered entity.
Researchers at Monica said Manus is the world's first fully autonomous AI agent. Unlike chatbots that require multiple inputs, it can complete complex tasks like sorting résumés, analyzing stocks, scraping data, and even building websites after a single prompt.
According to Manus' website, it outperforms OpenAI's Deep Research model on the GAIA benchmark, a tool for comparing models.
Following its debut last week, some users questioned whether Manus was built from scratch, suggesting it might be more of an AI wrapper leveraging other models, like Anthropic's Claude 3.5 Sonnet.
Yichao 'Peak' Ji, the cofounder of Manus, said in a post on X on Monday that the product uses Claude 3.5 Sonnet v1 and fine-tuned Alibaba's Qwen models.
Dean Ball, an AI policy researcher, said in a Sunday X post that it's "wrong to call Manus a 'DeepSeek moment'" — because it goes even further.
"DeepSeek was about replicating capabilities already achieved by American firms," Ball said. "Manus is actually advancing the frontier. The most sophisticated computer using AI now comes from a Chinese startup, full stop."
But others are calling it overhyped. TechCrunch's Kyle Wiggers and Pleias cofounder Alexander Doria said they found Manus prone to factual errors, execution failures, and endless loops during tests.
AI researcher Luiza Jarovsky also raised concerns about data privacy, asking where Manus stores its data and whether China has access.
For now, access to Manus requires an invitation.
5. Baidu's Ernie 4.5 and Ernie X1
China's competitor to Google, Baidu, this month released two newer versions of its AI model, Ernie.
Ernie X1 is a reasoning model that Baidu claimed could "deliver performance on par with DeepSeek R1 at only half the price."
The model has "stronger understanding, planning, reflection, and evolution capabilities," Baidu added.
Ernie 4.5 is a revamped version of the company's foundational model, which Baidu said "outperforms GPT-4.5 in multiple benchmarks while priced at just 1% of GPT-4.5."
With the release of these updated models, Baidu said it would integrate Ernie X1 and Ernie 4.5 into its search engine and wider product suite.
Ernie X1's reasoning model is significantly cheaper than OpenAI's o1, with prices just under 2% of OpenAI's model.
From June, Baidu's foundational model is set to be open-sourced, the company announced on X.
