.svg){kind=spacer}
Details
- OpenAI announced that its GPT-5, alongside an unnamed experimental reasoning model, jointly solved all 12 algorithmic tasks at the 2025 International Collegiate Programming Contest (ICPC) World Finals, equaling the top human team’s score.
- GPT-5 successfully handled eleven problems on its first try, while the toughest final problem was ultimately solved by OpenAI’s internal experimental model after GPT-5’s attempt fell short.
- The team deployed both models to generate multiple candidate solutions and used the experimental model to automatically select the highest-scoring answer, notably without any manual debugging, search optimization, or external tools.
- The ICPC World Finals is renowned for drawing the best university teams from over 100 countries to tackle a 12-problem algorithm challenge within a five-hour window, with perfect scores being exceptionally rare.
- OpenAI emphasized this milestone as a sign of significant progress since its 2024 o1-preview/o1-mini models, touting improved reasoning accuracy and lower costs with current state-of-the-art systems.
Impact
This flawless performance intensifies competition for Anthropic, Google DeepMind, and Meta, none of whom have replicated similar results in such a high-stakes contest. OpenAI’s feat could accelerate the adoption of AI coding assistants in the mainstream, raising concerns about job displacement and attracting regulatory attention. The approach also signals a shift toward ensembles and self-improving models, setting new benchmarks for the next wave of AI research and industry investment.
