Is AI Olympiad a Good Extracurricular?

Yes. If you plan to study computer science or artificial intelligence, a strong result in the USA AI Olympiad is now the competitive achievement that carries the most weight on your application. At Cosmic, we treat it as the gold standard for CS and AI applicants, a role the USA Computing Olympiad (USACO) held for years.

The USA AI Olympiad, officially the USA-North America AI Olympiad, is the national contest that selects the American team for the international olympiads in artificial intelligence. It was first held in 2024. In that short time, its field has filled with many of the strongest math and physics olympiad students in the country, and that concentration of talent is what makes a strong placement stand out.

Why AI Olympiad Has Replaced USACO as the Benchmark for CS and AI Applicants

USACO tests algorithmic programming, the skill of designing fast, correct procedures for well-defined puzzles under a time limit. For years it was the clearest competitive signal of computer science talent a high school student could offer.

The center of computer science has since moved toward machine learning, the branch of AI in which software learns patterns from data instead of following rules a programmer writes by hand. Machine learning now drives much of the research, hiring, and new coursework in CS departments, and a growing number of universities offer artificial intelligence as its own major.

The AI Olympiad tests that field directly. Every round combines the mathematics behind machine learning, mainly linear algebra, calculus, and probability, with the ability to build and train working models in Python. Its organizers describe the contest as equal parts math olympiad and coding olympiad, and a top result shows an admissions reader both halves at once.

USACO remains a strong indicator of programming ability. When you tell a college you want to study CS or AI, though, an AI Olympiad result speaks to the exact work you are asking to do, which is why we now place it first.

Why a New Competition Earned Its Reputation So Quickly

A competition's prestige comes from the students it attracts, and most new contests need years to build a strong field. The AI Olympiad skipped that wait. Many students already at the top of the AMC, the American Mathematics Competitions that lead to the USA Mathematical Olympiad, and the USA Physics Olympiad moved into AI and now prepare for this contest with the same intensity they bring to math and physics.

We see this directly in our work with STEM students. Students who spent years on math and physics olympiad problem sets now build their calendars around the AI Olympiad as well, with the same volume of practice problems and timed mock exams.

The fit is natural, because the contest rewards what olympiad training builds. Round 1 pairs hand calculation in linear algebra and calculus with code. The 2026 Round 2 asked students to derive, from probability theory, the formula that diffusion models (the method behind AI image generators) learn from, and to reconstruct a physical force law from measurements of a field. A student with years of olympiad practice arrives with most of the mathematics in place and adds programming and machine learning on top.

The numbers reflect that influx. In 2025, 76 students advanced from Round 1 to Round 2. In 2026, the organizers invited nearly 200.

The contest also feeds a fast-growing international olympiad. The International Olympiad in Artificial Intelligence (IOAI) drew more than 460 contestants from over 100 countries and territories to Astana, Kazakhstan, in August 2026, placing it in the size range of long-established science olympiads. Team USA brought home two gold, three silver, and three bronze medals. Closer to home, the 2026 Round 2 took place on the Harvard and MIT campuses with an information session run by MIT Admissions, and the June training camp was held at Harvard.

Because so many top students now compete, a strong placement means you outperformed a field that includes national-level math and physics olympians. That is what gives the result its weight.

How the USA AI Olympiad Is Structured

The contest runs in three stages each school year, and the 2027 cycle follows the same path as 2026.

Round 1 is open to any student. Registration is open now and closes on January 31, 2027, and the exam takes place on Friday, February 19, 2027, from noon to 3 p.m. Eastern. You take it at your school or at an approved site such as a university or library, under a proctor you arrange.

Round 1 has two parts. The first is a 60-minute section of multiple-choice and fill-in-the-blank questions answered inside a locked contest app. The second is a 75-minute coding section completed in Google Colab, a free browser-based tool for writing and running Python. You record your screen and your face for the full exam while monitoring software runs, and a single cheating finding brings a permanent ban. Top scores earn Honor Roll, High Honor Roll, or Distinguished Honor Roll recognition.

Round 2 is by invitation, extended to the top Round 1 scorers, and takes place in March or April. It uses the same two-part format with harder problems, some of which require a GPU, the kind of processor used to train neural networks. In 2026 it ran in person over two days at Harvard and MIT. Round 2 awards gold, silver, and bronze medals and honorable mentions, and its top scorers are named campers.

The camp is held in June. The 2026 camp ran June 8 to 12 at Harvard, with several team selection tests between lectures and talks from researchers and industry guests. Camp is open to U.S. citizens, permanent residents, and international students attending U.S. high schools.

Team USA is chosen at camp. Up to eight students represent the United States at the IOAI, held each summer and hosted by Singapore in 2027, and four compete at the International AI Olympiad (IAIO), held each winter.

What the USA AI Olympiad Asks

The syllabus covers the mathematics behind AI, Python programming, classical machine learning, and deep learning. The math centers on linear algebra, probability and statistics, multivariable calculus, and optimization. The programming uses NumPy, a Python library for fast calculations on large arrays of numbers, and PyTorch, the library most researchers use to build neural networks. A neural network is a model made of layers of simple numerical units whose connection strengths adjust as it learns.

The 2026 Round 1 shows what that looks like in practice. It had nine problems worth 300 points in total.

Several short questions checked core concepts, such as recognizing which task involves finding patterns in data that carries no labels, and how a model's errors shift as it grows more complex. Others required math by hand, including projecting one vector onto another, rewriting a table of numbers as a sum of simpler pieces, and differentiating a standard function. One problem then asked students to turn a calculus result into Python without loops or shortcut functions, which forces the whole calculation to run on entire arrays at once.

The largest problem, worth 90 points, dealt with word embeddings. An embedding represents each word as a list of 100 numbers, arranged so that words with related meanings have similar lists. Students measured the similarity between every pair of words in a provided text, found each word's closest partner, and proved properties of the resulting similarity table. They then compressed that table with singular value decomposition, a standard method for splitting a table of numbers into its most important components, and plotted how much accuracy each level of compression costs.

Other problems asked students to build a tiny neural network in PyTorch that decides whether a point lies inside a given triangle, choosing every internal number by hand with no training. Another took apart ResNet-50, a widely used image recognition network. Students counted its adjustable numbers, traced how an image's data changes shape as it passes through, and reused part of it for a new five-category task. A third showed five snapshots of what an image network detects at different depths and asked students to put them in order.

The round closed with an open-ended modeling task worth 50 points. Using a real medical dataset with 30 measurements per patient, students built a classifier based on k-nearest neighbors, a method that labels a new case by looking at the most similar known cases. Submissions were scored on data the students never saw, along with a written explanation of their reasoning.

Round 2 raises the level considerably. In 2026, its first problem asked students to derive and build linear attention, a faster version of the mechanism inside transformers, and then assemble a working transformer. Transformers are the architecture behind ChatGPT and most modern language AI. The remaining problems asked students to recover the location of a source and the law governing its force from measurements of a three-dimensional field, and to derive step by step the formula diffusion models learn from. Students also classified images of shapes when only a fifth of the images carried labels and pretrained models were banned, and separated data produced by four unknown kinds of functions while recovering each function's parameters.

The organizers' published design principles explain the approach. Every problem requires building from first principles and pairs mathematical reasoning with working code, and solutions that treat existing AI models as black boxes lose significant credit.

What Counts as a Competitive Placement

Each stage narrows the field, so the level you reach shows where you stand nationally.

A Distinguished Honor Roll score in Round 1 shows command of the full Round 1 syllabus. A Round 2 invitation places you in a group of roughly 200 students, most of them from the United States and Canada, and a Round 2 medal sets you apart within that group.

Camp is the result that separates an application at the most selective CS and AI programs. In 2025, only the top 12 Round 2 contestants advanced to camp, and we view a camp invitation as comparable to reaching the national training camps in math or physics. Team USA membership sits at the top.

For most students, reaching Round 2 is the right first target, with camp as the goal for a second or third attempt.

If you need help deciding which competitions and extracurriculars you should participate in to maximize your chances of getting into your dream school, schedule a free consultation with a college admissions expert today.