Math Circles Kids AI Era: Why Competition Still Develops Them

Math circles kids AI coverage keeps declaring obsolete are having an identity crisis this year, and they should not be. In July 2026, Huawei’s Celia and Xiaohongshu’s dots-note 3.0 were officially graded 42 out of 42 at the International Mathematical Olympiad in Shanghai. Among 666 human contestants, 7 scored the same.

If the point of olympiad training was to produce a person who can solve olympiad problems faster than a computer, that point is gone. It was never the point. Campbell and Walberg’s follow-up study of 345 adult Olympians, published in Roeper Review in 2011, found that 52% earned doctorates and the group had produced 8,629 publications. The training built researchers, not solvers.

Key Takeaways

• IMO 2026: 666 contestants from 117 countries, 55 gold medals at 29+ points, 7 perfect scores. Two AI systems were officially graded 42/42.
• Campbell and Walberg (2011), tracking 345 adult Olympians in Roeper Review: 52% earned doctorates, 8,629 publications produced.
• A math circle is a small group working on hard problems together, usually without grades or a set curriculum. That structure is the product.
• PISA 2025 found only 52% of OECD students finish tasks that have become boring, and perseverance fell in 32 to 40 systems since 2022. Sustained struggle is the scarce commodity.
• The OECD also found “the provision of study help, such as peer-to-peer tutoring, is associated with higher science performance.”

What math circles kids AI coverage overlooks actually do

A math circle is a small group of students who meet regularly to work on problems that do not appear on tests, guided by someone who knows the material well, usually without grades. The format came to the United States largely through Russian and Eastern European emigres and has spread through university math departments.

The distinguishing feature is not difficulty. It is that the problems are unfamiliar. In a typical school math class, a student sees a worked example and then does twenty problems shaped like it. In a math circle, nobody shows you the method, because the method is what you are supposed to invent.

That difference is the whole mechanism, and it matters more now, not less. An AI can solve a well-posed problem. It cannot sit with your child while they are stuck, watch what they try, and ask the question that unsticks them. And the experience of being stuck for forty minutes and then seeing it is what builds the tolerance for ambiguity that every technical career runs on.

The PISA 2025 data make this look urgent rather than optional. Across OECD countries, only 52% of students say they finish tasks that have become boring, the weakest item in the whole perseverance set. The share who apply extra effort when work becomes challenging fell in 32 systems since 2022. The share who finish what they start fell in 40. Sustained voluntary struggle is exactly what is declining, and it is exactly what a math circle sells.

Program comparison

Costs and formats change, so treat the dollar figures as rough and verify before committing. What I can characterize reliably is the structure of each option.

Program type	Format	What it develops best	Cost signal	Best fit
Local math circle	Small group, weekly, in person, usually university-hosted	Inventing methods; collaborative struggle	Often free or low cost	Any kid who likes hard problems
Art of Problem Solving (AoPS) courses	Structured online classes with instructors and forums	Systematic competition technique; written solutions	Paid, per course	Kids who want a curriculum and a community
AoPS books used independently	Self-study textbooks	Depth and independence	One-time book cost	Self-motivated readers
Summer math camps (e.g., residential programs)	Intensive weeks with peers at the same level	Immersion; finding people like you	Higher; financial aid often exists	Kids who feel isolated in their ability
School math team	Weekly practice, competition circuit	Speed, breadth, team identity	Usually free	Kids who like structure and teammates
Olympiad training programs	Selective, proof-focused	Writing rigorous proofs	Varies, often merit-based	Students already competing well
Online problem forums	Asynchronous discussion	Reading other people’s approaches	Free	Independent teenagers
One-on-one enrichment tutoring	Individual, tailored	Filling specific gaps	Highest per hour	Kids with a specific obstacle

The first row is where I would start, for a reason grounded in the OECD data rather than in preference: the report found that “the provision of study help, such as peer-to-peer tutoring, is associated with higher science performance.” Small-group problem work with peers is the closest thing on this list to that finding.

For families comparing costs against private tutoring, our analysis of Khan Academy versus a $150-an-hour tutor for math covers the evidence on paid instruction.

What the Olympiad follow-up research actually shows

I want to be careful here, because this is the strongest claim in the article and it needs its limits stated.

Campbell and Walberg published “Olympiad Studies: Competitions Provide Alternatives to Developing Talents That Serve National Interests” in Roeper Review (2011, volume 33, issue 1, pages 8 to 17). They analyzed data from 345 adult Olympians and found 52% earned doctorates, with the group accounting for 8,629 publications, many of them going on to positions in universities and research institutions.

The limits: this is a follow-up study of people who qualified for national olympiad teams, which means it is a study of an extraordinarily selected group. It cannot tell you that competition caused those outcomes, because the same students might have earned doctorates without ever attending a competition. Selection effects are enormous here.

What it does establish is a descriptive fact that matters for the AI question: the people who trained for these competitions went on to do research, publish, and build institutions. The skill developed was not “solve olympiad problems.” It was whatever transfers from that into a research career.

And it is the transfer, not the solving, that the 42/42 result leaves untouched. The arXiv paper “From Solvers to Research” (July 2026, 19 authors including Terence Tao) argues mathematicians retain “aesthetic judgment about which problems merit investigation, verification of solutions and proofs, and directing overall research strategy.” Olympiad training is closer practice for those three than any homework set.

The honest case against pushing competition

I do not want to sell this uncritically, so here is the other side.

Competition suits some kids and damages others. Timed, ranked, high-pressure formats can turn a child who loved mathematics into one who associates it with anxiety. Our piece on math anxiety and what brain scans reveal covers the mechanism. If your kid dreads it, the format is wrong even if the mathematics is right.

The pipeline can become the point. When a family treats math circle as a college-application input, the thing that makes it work, voluntary engagement with a hard problem, gets replaced by compliance. The PISA finding that engagement and performance fell together is a warning about exactly this.

Access is uneven. Math circles cluster near universities. Camps cost money and require travel. This is a real equity problem, and the honest answer is that a strong local option beats a prestigious distant one you cannot sustain.

Speed is overweighted. Many competitions reward fast pattern recognition, which is not the same as mathematical depth. Some excellent mathematicians were mediocre competitors. Terence Tao’s own view, as reported in coverage of AI reasoning, distinguishes exhaustive exploration of defined spaces from conceptual breakthroughs; competitions test more of the former.

What to actually do at home

Find the local option before the prestigious one

A free weekly circle at a nearby university, attended consistently for two years, does more than one expensive summer program. Search for “math circle” plus your city, and check the nearest university mathematics department.

Protect the stuck time

The developmental value is in the forty minutes of not knowing. If you rescue, or if your kid looks up the answer, the transaction completes with nothing built. This is where AI creates a new hazard: the stuck time is now one prompt away from ending.

Ask for the attempt, not the answer

“What did you try?” is the only question worth asking about a competition problem. It rewards the process the format is designed to build and it does not punish failure.

Use AI as a checker, never as a solver

If your kid uses a model on a hard problem, the rule is that they attempt first and the model checks after. Reversing that order destroys the entire point of the activity. Our piece on what a 50% AI success rate teaches about reliability makes the case for the checking habit specifically.

What not to do

Do not use the 42/42 result as a reason to drop competition, and do not use it as a reason to push harder. It changed what machines can do, not what a fourteen-year-old gets from spending an hour stuck on a problem and then solving it.

What to Watch For Over the Next 3 Months

• Week 4: Your child spends 20 or more minutes on a single problem without asking for help or looking it up. That is the capacity being measured.
• Month 2 red flags: Dread before a session, or asking whether the competition “still counts” now that AI can do it. The second question deserves a real answer, not reassurance.
• Month 3 self-check: Ask your kid to describe a problem they failed to solve and what they tried. If they can do that without embarrassment, the culture around it is healthy.

Frequently Asked Questions

Do math competitions still matter now that AI scores 42/42?

The case never rested on machines being unable to solve the problems. Campbell and Walberg’s study of 345 adult Olympians found 52% earned doctorates and the group produced 8,629 publications, which speaks to what the training builds in a person rather than to who solves fastest.

What exactly is a math circle?

A small group of students meeting regularly to work on unfamiliar problems, guided by someone who knows the material, usually without grades or a fixed curriculum. The defining feature is that nobody shows you the method, because inventing the method is the exercise.

Is AoPS worth the money?

It depends on what your child needs. AoPS provides structure, a curriculum, and a peer community, which suits students who want a path. A free local circle provides in-person collaborative struggle, which the OECD’s finding on peer-to-peer study help points toward. Neither dominates.

Can competition math hurt a kid?

Yes, for some kids. Timed, ranked formats can convert enthusiasm into anxiety. If your child dreads sessions, the format is a poor fit even if their mathematical ability is strong. Non-competitive circles and self-study are legitimate alternatives.

How many students actually reach the top?

At IMO 2026, 55 of 666 contestants earned gold (29 points or more) and 7 scored a perfect 42. Those contestants were already national-team selections. The overwhelming majority of students who benefit from competition math never reach that level, and the benefits do not require it.

---

About the author

Ricky Flores is the founder of HiWave Makers and an electrical engineer with 15+ years of experience building consumer technology at Apple, Samsung, and Texas Instruments. He writes about how kids learn to build, think, and create in a tech-saturated world. Read more at hiwavemakers.com.

---

Sources

1. Campbell, J. R., & Walberg, H. J. (2011). “Olympiad Studies: Competitions Provide Alternatives to Developing Talents That Serve National Interests.” Roeper Review, 33(1), 8–17. https://www.tandfonline.com/doi/full/10.1080/02783193.2011.530202
2. International Mathematical Olympiad. (2026). “IMO 2026, Shanghai, July 10–21: results.” https://www.imo-official.org/editions/2026/
3. OECD. (2026). “PISA 2025 Results (Volume I): Future-Ready Students,” Chapters 3 and 4. OECD Publishing. https://www.oecd.org/en/publications/pisa-2025-results-volume-i_73451bc5-en.html
4. Jiang, E., Liang, X., et al. (with Tao, T.). (2026, July). “From Solvers to Research: Large Language Model-Driven Formal Mathematics at the Research Frontier.” arXiv:2607.07779. https://arxiv.org/pdf/2607.07779
5. Chen, R. (2026, July 22). “World’s first AI model to earn perfect score at maths olympiad comes from China’s RedNote.” South China Morning Post. https://www.scmp.com/tech/article/3361482/worlds-first-ai-model-earn-perfect-score-maths-olympiad-comes-chinas-rednote
6. Nickow, A., Oreopoulos, P., & Quan, V. (2020). “The Impressive Effects of Tutoring on PreK-12 Learning: A Systematic Review and Meta-Analysis of the Experimental Evidence.” NBER Working Paper 27476. https://www.nber.org/papers/w27476
7. Campbell, J. R., et al. “Mathematics and Science Olympiad Studies.” ERIC. https://files.eric.ed.gov/fulltext/EJ1301497.pdf