A professor's blunt warning has ignited a viral debate this semester. "If you don't have paid tokens in hand, you should drop out immediately," said Jiang Yanran, an associate professor at Nanjing University, in a remark that swept social media and climbed the Zhihu hot list. His statement, of course, requires nuance. He was using hyperbole to drive home a point: tokens have become the consumable supplies of the AI era, and every student must possess and master them.

While Professor Jiang was addressing students, Stanford Professor Fei-Fei Li sounded a similar alarm to university presidents across the Pacific. Reposting OpenAI's latest efficiency data, she wrote that the R&D world is splitting into "token-rich research" and "token-poor research". At top AI companies and emerging research institutions, researcher capabilities are being amplified by abundant AI tokens. The future of university research, she warned, deserves serious reflection from every university leader.

Though addressing different audiences, both are pointing to the same underlying shift: AI is redefining research resources. A well-equipped lab once meant GPUs, servers, and experimental equipment. Now, literature retrieval, experiment design, failure analysis, and exploring new directions all depend on tokens. Some argue tokens are the pen and paper of the AI era, a fundamental input for scientific production.

Industry has already voted with its feet. Top AI firms allocate tokens to research teams almost on demand, and issuing token allowances to engineers and researchers is becoming standard practice. The unspoken consensus: giving researchers ample tokens is giving them the chance to produce results. This consensus, however, has yet to reach the broader student population.

That is beginning to change. On September 21, Tencent officially launched its second Qingyun Scholarship, with each recipient receiving a token allowance valued at RMB 300,000. This time, tokens are being handed directly to students. The scholarship focuses on addressing practical research challenges and pain points, and was successfully held last year, with Tencent's Chief AI Scientist and lead reviewer of the scholarship, Yao Shunyu, presenting awards to 15 outstanding young researchers at the ceremony. Application link: https://join.qq.com/scholarshipapply.html

What does RMB 300,000 in tokens actually mean for a student? To understand the impact, you first need to know how students currently spend tokens and how scarce they are. We spoke with several doctoral candidates to find out.

In university labs, token scarcity is a daily reality. The PhD students we interviewed reported monthly token expenses ranging from a few hundred to several thousand yuan. Some can claim reimbursement through their research groups; others must cover costs personally. Social media posts suggest the latter is more common. Regardless of who pays, one thing is universal: every token must be carefully budgeted. As a result, tasks that could be delegated to AI often end up being done manually to conserve tokens — breaking large tasks into smaller ones, optimizing workflows by hand, and limiting context lengths. Or researchers test with cheaper, more conservative models before deciding whether to deploy stronger ones. These workarounds are acceptable, but the cost is real: increased time investment and, more critically, a budget cap on trial and error.

So what if the budget suddenly vanished as a constraint? We asked the same students a follow-up: if they had a huge token reserve, what would they do first? Their answers were strikingly aligned. They would stop treating token costs as a research constraint — more boldly involving models in idea exploration, feeding massive case traces to AI for failure analysis, generating code prototypes more frequently, and running multiple experimental paths. In short, they don't lack ideas; they lack the resources to test them quickly.

So what would RMB 300,000 in tokens mean for them? The math gives a rough picture. Using current API pricing of mainstream frontier models, and assuming a 4:1 input-to-output ratio, RMB 300,000 could buy several billion mixed tokens at the top tier. For more cost-effective options, such as DeepSeek V4.1 Flash, the total could reach into the hundreds of billions. But billions are still abstract. Framing it in research tasks helps build intuition. The Stanford AI Index 2026 tracked performance of research agents on AstaBench, which contains over 2,400 research tasks covering literature understanding, code execution, data analysis, and end-to-end scientific discovery. Among 57 agents tested, the top performer averaged about USD 3.4 per benchmark problem, with most agents coming in under USD 1 per problem. By those figures, RMB 300,000 could cover roughly 13,000 to 45,000 research benchmark questions.

A 2026 study on Terminal-Bench illustrates the opposite extreme: most agents spend under 20 minutes on tasks, but some complex tasks see agents running for nearly two hours, firing off hundreds of API calls, with a single task potentially consuming close to 100 million tokens. In other words, if you make each task heavy enough, RMB 300,000 could still run dozens to thousands of long-chain, deep-running agent experiments. Different calculation methods yield very different numbers, but by any measure, this represents an exceptionally "luxurious" AI research assistant.

On a personal level, the figure hits home. Those doctoral students, spending hundreds to thousands of yuan monthly and pinching every token, estimated that this allowance could provide them with one to three years of "token freedom" at their ideal usage intensity. In other words, what RMB 300,000 in tokens buys isn't just API calls — it's the scarcest commodity during a PhD: the freedom to make mistakes.

Beyond the tokens, what else does the scholarship offer? The RMB 300,000 token allowance addresses a major challenge, but it isn't the full picture. Recognizing the other expenses and financial pressures students may face, this year's Qingyun Scholarship also grants each recipient RMB 200,000 in cash, with no restrictions on how it can be used. Last year's recipients reported using the funds for API and token costs (now covered separately), and for covering academic conference travel. Previously, attending cross-city conferences required careful budgeting; now it's far more manageable. Some even brought lab mates and junior students along, giving non-first-authors a chance to attend top conferences in person.

Beyond the token innovation, the scholarship's scope has expanded. Last year it supported 15 recipients; this year, 20. Eligibility has also broadened from mainland China, Hong Kong, Macau, and Taiwan to global applicants: master's and doctoral students of Chinese nationality enrolled at universities worldwide in computer science, AI, and related interdisciplinary fields, with expected graduation dates of January 2028 or later. The timeline is straightforward: applications run from September to October (deadline October 25), reviews from November to December, and awards in January.

Yao Shunyu, lead reviewer, also shared his expectations. Qingyun seeks young people with genuine research potential, innovative thinking, and long-term growth prospects. In his view, real research and innovation often require confronting questions without answers — continuously trying, failing, and exploring to find something truly worth pursuing. He values intrinsic motivation over papers and awards. "When you are genuinely interested in a problem, willing to think about it and explore it for years, even decades, that drive matters more than any paper or award."

This explains why the scholarship goes beyond money and tokens. Young people with genuine self-drive can be identified through the selection process, but converting that drive into sustained research momentum requires more than individual effort — it requires intangible "external forces." Often, that force comes from connection. During last year's poster sessions and closed-door discussions, recipients could directly discuss their research with Tencent technical experts and meet peers from different universities and fields. Crucially, these connections didn't end with the event. Deng Yangtao from Chinese University of Hong Kong noted that after winning, he built ongoing relationships with industry mentors and peers, with collaboration opportunities continuing to emerge. Song Liyang, working in computational biology and statistical genetics, had a similar experience: he met experts and friends in the life sciences industry, some of whom he remains in contact with today. For those in fundamental science, these exchanges offer a direct benefit: a stream of feedback from real industrial environments, in addition to academic peers.

These changes are harder to quantify than RMB 200,000 in cash or RMB 300,000 in tokens, but they may yield returns over a much longer horizon. Research certainly needs resources, but it also needs feedback, benchmarks, and collaborators. For young researchers just establishing their direction, the ability to step outside their academic circle and engage with different problems and perspectives can shape their next steps. The Qingyun Scholarship, then, offers two things: one that removes immediate constraints, giving young researchers more room to experiment; the other that connects them to a broader research network, ensuring that conversations continue long after the award ceremony.

A recent headline has captured attention: two American high school students, Aayush Bathija and Prince Rohatgi, working under the mentorship of UCLA mathematics postdoctoral fellow Daniel Soskin and using AI tools like Claude Opus 5 and GPT-5.6 Sol, advanced a mathematical problem previously explored by Fields Medalist June Huh and others. They generalized results on Lorentzian polynomial coefficient ratios from a special quadratic case to a more general setting. The work was published as a paper on arXiv. Both students are members of UCLA's ORMC math circle. What's remarkable isn't just that high schoolers produced mathematical research — it's a reminder that when a young person has access to powerful AI tools, good mentors, and a continuous peer community, the barriers of age and accumulated knowledge are being rewritten.

We posed a similar question to a doctoral student: given RMB 300,000 in tokens, would young researchers achieve breakthrough results? His answer was direct: "Yes. There's no knowledge that can't be learned now. With AI, most knowledge is accessible, and many parts of research can be guided and supported by AI." Of course, AI tools address capability and efficiency — they aren't everything. Having someone to tell you whether a problem is worth pursuing, peers to exchange ideas with, and people from different fields offering fresh feedback — these connections equally shape how far a piece of research can go. This is what makes the Qingyun Scholarship's mission, "for more worthwhile exploration," particularly relevant today: providing both resources and connections, so that those with genuine curiosity and intrinsic drive can go further. The AI-native generation of researchers now has unprecedented tools. The more compelling question ahead is: with tools, resources, and connections becoming more accessible, which previously untouchable problems will they choose to pursue?