Reported by NUPIAO
Recently, a screenshot of a DeepSeek intern’s salary went viral on social media. It showed a daily pre-tax pay of 5,500 RMB—that’s over 120,000 RMB a month based on a 22-day work schedule. This stands in stark contrast to the typical 500 to 1,000 RMB per day for AI interns in the industry.
Some netizens pointed out that such eye-watering pay is reserved for a tiny handful of top graduates from Tsinghua’s Yao Class—a “VIP price tag” that big tech companies slap on the cream of the AI crop.
While DeepSeek hasn’t confirmed the rumor, the Tsinghua Yao Class’s reputation in AI is already legendary. Some go so far as to claim, “The Tsinghua Yao Class props up half of the global AI industry.”
In investment circles, if word gets out that a Yao Class alum is launching a startup, VCs are all over it. For two decades now, this tiny program—admitting just over 50 students each year—has churned out founders in autonomous driving, computer vision, large models, graphics software, medical AI, and blockchain.
Take Lou Tiancheng from the very first Yao Class cohort. He was recommended for admission to Tsinghua’s CS department in 2004, then went on to get his PhD at the university’s Theory of Computation Center in 2008. In 2016, he co-founded Pony.ai with Peng Jun, focusing on Level 4 autonomous driving. The company went public on Nasdaq in 2024, becoming the world’s first “Robotaxi IPO.”
Megvii’s founders—Yin Qi, Tang Wenbin, and Yang Mu—are all Yao Class alumni, famously dubbed the “Three Musketeers of Yao Class.” In 2006, Yin Qi scored over 680 on his college entrance exam and got into Tsinghua’s Automation program, later being selected for the Yao Class. He went on to pursue a PhD at Columbia University in New York in 2011, and that same year, he co-founded Megvii with Tang Wenbin and Yang Mu. The company became one of China’s “AI Four Little Dragons” by commercializing facial recognition tech. Today, Yin Qi serves as Chairman of Stepfun.
Some Yao Class grads have landed gigs at the world’s top tech firms. Yao Shunyu, from the 2015 cohort, worked at OpenAI and now serves as Tencent’s Chief AI Scientist. Meanwhile, xAI, Google DeepMind, and Meta AI have all snapped up multiple Yao Class alums for cutting-edge research in underlying algorithms, large model theory, and multimodal generation.
According to Tsinghua’s official website, as of 2024, nearly 60 Yao Class graduates are teaching at universities worldwide—over 30 in China—with seven returning to Tsinghua’s Institute for Interdisciplinary Information Sciences.
Behind all these AI stars is the mastermind of the Yao Class: Yao Qizhi. Back in 1967, he got his bachelor’s in physics from National Taiwan University, then went on to earn a PhD in physics from Harvard and another in computer science from the University of Illinois. Over the next three decades, he taught at four of the world’s best universities: MIT, Stanford, UC Berkeley, and Princeton.
In 2000, Yao Qizhi won the Turing Award for his foundational work in communication complexity and modern cryptography. At that point, he had a tenured position abroad, a fat paycheck, and global academic fame. Yet at 57, he decided to quit his Princeton professorship and return to China in 2004.
The turning point came from a transatlantic phone call with Academician Yang Zhenning. Back then, algorithm and complexity research in China was practically nonexistent. Yang urged Yao to come back and nurture homegrown computer talent.
In 2005, Yao Qizhi officially founded the Tsinghua Xuetang Computer Science Experimental Class—better known as the Yao Class. Tsinghua admits over 3,000 freshmen each year, but the Yao Class only takes about 50. The students are typically the top three in their province’s college entrance exams, or gold medalists in math, physics, and informatics Olympiads, recruited through recommendations and a second round of selections on campus.
Unlike the standard CS curriculum, the Yao Class uses a step-by-step course structure. The first two years focus on computer fundamentals, while from junior year onward, they switch to original MIT textbooks. Yao Qizhi personally interviews every instructor and even invites Turing Award and Gödel Prize winners to teach in the classroom.
In the Yao Class, the coursework is often way harder than regular undergrad classes. One student described the pace as “building a skyscraper starting from the sixth floor”—skipping basic coding and diving straight into deep computational theory and system architecture. And it’s not all about grades. As Megvii co-founder Yang Mu once put it, “In the Yao Class, solving problems no one has solved before is just another Tuesday.”
As of December 2025, the Yao Class has trained nearly 1,200 students, with close to 750 having graduated. Among them are 9 winners of Tsinghua’s “Undergraduate Special Prize” and 6 recipients of the “Sloan Research Fellowship.” Yao Class students have published over 600 papers and delivered nearly 330 presentations at top international conferences in computer science, AI, and quantum information.
As an elite talent incubator created by Yao Qizhi after his return to China, the Yao Class has, with a tiny size, churned out top-tier algorithm engineers active in core global AI sectors, filling a gap in China’s computer science theory education.
But some voices caution against over-mythologizing the Tsinghua Yao Class. It’s a special pilot program for serving top-tier tech competition, not a reflection of China’s overall CS education. The industry needs to keep a balanced view of its achievements and limitations. Balancing elite training with inclusive education, and guiding talent to straddle both industrial application and long-term basic research—that’s what real education should be about.
As Yao Qizhi recently said at the 2026 World Artificial Intelligence Conference, “The old order won’t work anymore. Whether you’re from MIT or Tsinghua, five years from now you might not be standing at the top of the world. It all depends on how hard teachers and students innovate. Top schools might even be at a disadvantage because they’re too comfortable with their success models. Meanwhile, ambitious startups, driven by their eagerness to seize opportunities, might end up with outstanding results.”
He reminded everyone that as long as you’re in the AI game, you have to keep adapting to the world.
