Jakarta, INTI – The growing need for increasingly complex algorithmic solutions in the digital era often pushes humans to the limits of their capabilities. This challenge was evident at the 2025 International Collegiate Programming Contest (ICPC), the world’s most prestigious programming competition, held on Sunday, September, 21, 2025. The event, which brought together 139 universities from 103 countries, delivered a major surprise: two artificial intelligence models OpenAI’s GPT-5 and Google DeepMind’s Gemini 2.5 Deep Think were tested on the same problems as human participants. The results were astonishing: GPT-5 achieved a perfect score by solving all 12 problems, while Gemini solved 10 problems, including one that no human team could crack. This phenomenon demonstrates how AI can deliver solutions once thought impossible.
AI Meets the Challenge of a Global Contest
Although not competing officially, the ICPC organizers provided GPT-5 and Gemini with the exact same problem set as human contestants, under the same five-hour time limit. GPT-5 completed all problems with a flawless score an achievement beyond even the strongest university teams. OpenAI emphasized that the model had not been specially trained on ICPC problems, yet it consistently found correct answers in nearly every initial attempt.
Meanwhile, Google’s Gemini 2.5 Deep Think solved 10 problems, including one involving liquid distribution that stumped every human team. Using a minimax theorem based approach, Gemini calculated the optimal flows and discovered the most efficient distribution pattern.
The Competition and Human Comparison
At ICPC 2025, participants were challenged to solve 12 algorithmic problems within five hours. Leading universities such as St. Petersburg, Tokyo, Jiaotong Beijing, and Tsinghua earned gold medals with 11 problems solved, while Harvard and MIT took silver as the best U.S. representatives. However, no human team managed a perfect score like GPT-5.
Google highlighted that Gemini’s performance illustrates AI’s potential to complement human ability, especially in cases involving complex calculations and multi-scenario analysis.
Implications for Academia and Technology
This achievement is more than symbolic. Both OpenAI and Google consider the ICPC results proof that large language models (LLMs) can adapt to high-level academic challenges and even surpass human performance. OpenAI had previously tested GPT in the International Mathematical Olympiad (IMO) and the International Olympiad in Informatics (IOI), demonstrating consistent performance across multiple disciplines.
Conclusion
The participation of ChatGPT (GPT-5) and Gemini 2.5 in ICPC 2025 marks a turning point in the world of programming. AI is no longer just an auxiliary tool, but a contender capable of standing alongside and even surpassing human capacity in solving intellectual challenges. If this trend continues, academia and industry alike must prepare for a new era of collaboration between humans and artificial intelligence.
Read More:China & ASEAN Launch AI Cooperation Center in Nanning