News

OpenAI beats Google and humans: historic ICPC 2025 record

Article Highlights:
  • OpenAI solved all 12 ICPC 2025 problems
  • Google Deepmind solved 10 problems, OpenAI 12
  • GPT-5 and experimental model collaborated for the win
  • No model was specifically trained for ICPC
  • The result highlights advanced AI model superiority
  • OpenAI aims for systems that can discover new knowledge
  • Competition was judged officially
  • Success follows similar results at Math and Informatics Olympiads
OpenAI beats Google and humans: historic ICPC 2025 record

Introduction

OpenAI set a historic milestone at the International Collegiate Programming Contest (ICPC) 2025, outperforming top university students and Google Deepmind. Its AI solved all 12 problems, establishing a new benchmark in the competition between artificial intelligence and human talent.

Context

The ICPC is the world’s most prestigious collegiate programming contest. In 2025, OpenAI’s system participated under the same conditions as human teams, receiving the problem set in PDF format and facing a five-hour time limit. Solutions were judged officially, with no special adjustments.

Direct answer

OpenAI solved every ICPC 2025 problem, surpassing both Google Deepmind and top university teams.

Comparison with Google Deepmind

Days earlier, Google Deepmind announced that Gemini 2.5 Deep Think reached gold-medal level, solving 10 out of 12 problems, including one no human team solved. However, OpenAI achieved a flawless result, solving all 12 problems and surpassing Deepmind.

OpenAI’s solution: GPT-5 and experimental model

The OpenAI system consisted of general-purpose reasoning models, not specifically trained for ICPC. GPT-5 produced correct answers for 11 problems, while the experimental model selected which solutions to submit. The final, hardest problem was solved by the experimental model after nine attempts, highlighting advanced reasoning capabilities.

  • GPT-5 solved 11 problems
  • The experimental model solved the twelfth
  • No model was specifically trained for ICPC

Implications and future outlook

OpenAI sees this achievement as part of steady progress. Its models have already reached gold-level results at the International Mathematical Olympiad and the International Olympiad in Informatics. According to Mostafa Rohaninejad, the next goal is to build systems capable of discovering new knowledge.

“The next milestone will be building systems that can discover new knowledge, the true leap forward.”

Mostafa Rohaninejad, OpenAI

 

FAQ

  • How did OpenAI solve every ICPC problem?

    It used GPT-5 and an experimental model, without specific training for the contest.

  • What’s the difference between OpenAI and Google Deepmind at ICPC?

    OpenAI solved 12 out of 12 problems, Deepmind solved 10 out of 12.

  • Why is OpenAI’s ICPC result important for AI research?

    It shows advanced AI models can outperform both humans and other AI systems.

  • What limitations did GPT-5 show during the competition?

    GPT-5 failed on the hardest problem, which was solved by the experimental model.

  • Did OpenAI use models specifically trained for ICPC?

    No, the models were general-purpose reasoning systems.

  • What are the future prospects for AI research after ICPC?

    OpenAI aims for systems that can discover new knowledge.

Introduction OpenAI set a historic milestone at the International Collegiate Programming Contest (ICPC) 2025, outperforming top university students and Evol Magazine
Tag:
OpenAI