IOAI² Inaugural Edition Review: Autonomous AI Meets the International Olympiad in AI Challenge

What can AI achieve when challenged with the same problems as talented young students?

The International Olympiad in Artificial Intelligence (IOAI), at its 2026 edition, launched the inaugural IOAI²: AI Models Track turning the IOAI tasks into a new test of autonomous AI.

IOAI² brought together 15 autonomous AI modeling systems from 14 founding organizations around the world, from leading university research labs to startups and independent AI labs. The track winner was Synthetic Sciences, a Y Combinator-funded AI lab co-founded by a student participant at IOAI 2025. 

Each system had two hours to explore unfamiliar data, develop a solution and improve it autonomously, without human guidance. The systems tackled the same six tasks as students participating in IOAI 2026, where 108 countries and territories were represented.

On one of the six IOAI 2026 tasks, the best student score exceeded the best AI score, and on two tasks the best student score was within 7% of  the best AI score. Across all six tasks, however, all 15 autonomous AI systems achieved overall scores at or above the student gold-medal threshold, and 13 exceeded the highest student total. The two groups competed under different conditions, so these figures do not provide a direct measure of AI versus human ability under equal settings. In an unexpected twist, several AI agents attempted to exploit loopholes in the competition rules or Kaggle platform, which the IOAI Jury detected and addressed.

This inaugural IOAI²: AI Models Track edition review presents the competition format, participating organizations, results, costs of participation and proctoring, Jury precedents and lessons for future editions. This offers other scientific olympiads a proposed approach for evaluating fully autonomous AI systems under their own governance.

What is IOAI and IOAI²?

The International Olympiad in Artificial Intelligence (IOAI) is an annual competition and education initiative for school students. Its Individual Contest asks them to solve AI modeling problems under a fixed time limit and controlled conditions. IOAI was founded in 2024 in Bulgaria, the second edition was held in 2025 in China and the third edition in 2026 was held in Astana, Kazakhstan, where 108 countries and territories were represented.

In 2026, IOAI inaugurated IOAI²: AI Models Track. IOAI² is a competition for autonomous AI systems that brought together teams from leading university research labs worldwide and emerging AI startups, including ventures co-founded by IOAI alumni from 2025 and 2026. Its inaugural edition ran in parallel with the on-site student Individual Contest at IOAI 2026, with AI systems evaluated on the same six tasks following as closely as possible the Contest rules of IOAI.

IOAI²: AI Models Track connects the Olympiad with organizations building autonomous AI systems and gives those systems a common competition on previously unseen AI modeling tasks. Participating organizations use their own models, tools and execution infrastructure. The IOAI International Scientific Committee defines the tasks, scoring rules, rankings and awards, while the IOAI Jury proctored the competition on Kaggle and reviewed compliance with the published rules.

IOAI² supports independently reviewed evaluations of fully autonomous AI modeling systems.

The track examines how well these systems solve novel AI modeling problems within a two-hour competition window and without human guidance. It evaluates several complementary abilities: understanding an unfamiliar problem and data, managing time, selecting and running experiments under resource constraints, submitting code through a CLI (Command Line Interface), interpreting feedback, and improving a solution through repeated iteration. Together, these abilities measure sustained, autonomous problem-solving on olympiad-level AI modeling tasks.

Official Statement: