AI outscored the best humans at two world coding contests in 2026, a year after falling short
At AtCoder's world finals and the International Olympiad in Informatics, AI scored above every human in 2026. Both runs were unofficial.
Competitive programming contests give people a few hours to solve tricky coding puzzles. The International Olympiad in Informatics (IOI) is the top contest for school students. AtCoder, a Japanese contest site, runs its own World Tour Finals.
2025. At the IOI, OpenAI’s models scored at gold-medal level under contest rules, beating all but 5 of 330 students, up from the 49th percentile in 2024. At AtCoder’s World Tour Finals Heuristic contest (10 hours, with answers scored by how good they are rather than right or wrong), human champion Przemysław “Psyho” Dębiak beat OpenAI’s model. It finished second after leading for most of the contest.
2026. At AtCoder’s World Tour Finals in Tokyo in July, an OpenAI agent beat all 12 human finalists in the heuristic exhibition. In the algorithm exhibition it solved all five problems for 8,300 points, against 4,300 for the best human; no human solved two of them. At IOI 2026 in August, NVIDIA reports that its Nemotron-3-Ultra model scored 535.4 of 600 under the same time and submission limits as the students. The top student scored 498.27, and the gold-medal cutoff was 361.12.
What it means
On short, well-defined coding puzzles under time pressure, the best AI systems now outscore the best people in the world. That says little about everyday software work, which involves long projects, unclear requirements and teams. Both 2026 results are also unofficial, and one is NVIDIA’s own report.
Caveats
- The AtCoder runs were sponsor exhibitions, not official rankings.
- NVIDIA’s IOI result is self-reported in a preprint, not adjudicated by the IOI, with no compute disclosed.
- OpenAI’s 2026 AtCoder model was unreleased.
- These contests test short, well-specified problems, not long-term software engineering.