Just now, OpenAI and Gemini jointly won the gold medal at ICPC 2025. OpenAI achieved a perfect score, crushing all the competitors.

It's crazy!

Just now, Google and OpenAI both won the gold medals at the ICPC. In particular, OpenAI even got a perfect score!

The full name of ICPC is the International Collegiate Programming Contest, which is one of the most prestigious programming competitions in the world!

The rule is to solve more than a dozen extremely complex programming and algorithm problems within five hours!

Finally, Gemini successfully solved 10 out of 12 problems and won the gold medal.

OpenAI solved all the problems correctly, got a perfect score, and won the gold medal!

What about humans?

Among the 139 human participating teams, only 3 teams achieved the same result as Gemini (10/12), and no human team got a perfect score.

The only Chinese team that tied with Gemini was Beijing Jiaotong University. We also conducted in - depth reports at the moment when the results of the ICPC Global Finals were announced, analyzing how this strongest Chinese team was formed.

Beating Harvard and MIT! Beijing Jiaotong University and Tsinghua University won the gold medals at the 2025 International Collegiate Programming Contest!

In particular, Google also specifically mentioned that all human teams failed to solve Problem C, while Google's Gemini successfully solved it within half an hour!

OpenAI solved all the problems and got a perfect score!

It's really a shocking moment, a historic night. AI has completely surpassed humans in the top - level programming competition!

Gemini Solved Problem C That All Participating Human Teams Failed to Solve

Google's official account announced that the advanced version of Gemini 2.5 Deep Think achieved a gold - level result at the ICPC 2025.

According to Google, it didn't create a brand - new trained model for the ICPC as it did for the International Mathematical Olympiad (IMO) earlier this year.

The Gemini 2.5 participating in the ICPC is the same model as the one we use in the Gemini application.

However, it has been enhanced to be able to think continuously for five hours!

In the ICPC competition, only correct answers can score points, and the time required to get the answer will affect the final score.

After the competition started, Gemini quickly climbed to the top few positions, correctly solving 8 problems in just 45 minutes.

Gemini finally answered 10 questions correctly and ranked second among the university teams.

Google specifically mentioned that during the competition, Gemini successfully solved Problem C, which no other human team could solve.

Problem C requires finding a solution to distribute liquid through an interconnected pipeline network to a group of reservoirs. The goal is to find a pipeline configuration that fills all the reservoirs with liquid as soon as possible.

Since each pipeline can be open, closed, or even partially open, the number of possible configurations is infinite, so it's very difficult to find the optimal configuration.

Gemini's solution is to assume that each reservoir has a priority value, which allows the model to use the dynamic programming algorithm to find the most efficient configuration.

After 30 minutes of repeated thinking, Gemini used the nested ternary search to determine the correct answer.

We asked Cursor to evaluate the answer announced by Google, and Cursor also thought that Google's solution method was very perfect.

Google has published all the other problems on GitHub. If you're interested, you can have a try!

https://github.com/google-deepmind/gemini_icpc2025

The Secret Behind OpenAI's AK at the ICPC

Gemini has proven its powerful strength by solving the extremely difficult Problem C that made all humans fail.

However, there's still a master! OpenAI's reasoning model directly cleared all 12 problems in the competition and achieved AK (All Kill, meaning solving all the problems) at the ICPC!

The research team behind it also posted 8 tweets in a row to celebrate this historic moment.

It's worth noting that 11 of the problems were solved directly using GPT - 5, which is available to everyone. Only the most difficult problem (it may refer to Problem C that stumped all humans, or it may refer to the last Problem L, or it may be other problems that are more difficult for large models) was solved using a mysterious experimental reasoning model with an unannounced model number.

Could this mysterious experimental model be the next - generation brand - new reasoning model?

In the last tweet, several key behind - the - scenes researchers were also @ mentioned and made their appearances.

After querying their identities, we learned that they are all key researchers of the reasoning models after o1.

One of them is Borys Minaiev, a member of the world - champion team at the ICPC in 2015.

His well - known ID in the algorithm field is: qwerty787788.

Among the three members of that championship team, there's another more well - known legendary figure, tourist, the strongest king in the algorithm field and the world's number one!

In addition, Jakub Pachocki, the Chief Scientist of OpenAI, is also a world - gold medalist at the ICPC. After the competition, he also retweeted the tweet and expressed warm congratulations on the excellent performance of the team.

The participation of many world - top ICPC players in OpenAI is a strong support for OpenAI to finally train a large model that can achieve AK at the ICPC.

A Historic Night

Last time when OpenAI announced the IMO results ahead of Google, there was a small stir.

Both Google and OpenAI wanted to compete to be the first AI to win a competition at the IMO level.

This time, their simultaneous announcements are of great significance!

Dr. BILL POUCHER, the Global Executive Director of the ICPC, believes that:

Gemini's successful entry into this field and achieving gold - level results marks a crucial moment in defining the artificial intelligence tools and academic standards required for the next generation.

AI winning the gold medal at the ICPC has direct and practical significance for software development.

This shows that artificial intelligence can be a real problem - solving partner for programmers.

Whether it's Google's CEO Sundar Pichai or OpenAI's Chief Scientist, they both personally supported their respective AIs!

In addition to programming and mathematics, AI's powerful abstract reasoning ability can be applied to many scientific and engineering fields, such as designing new drugs or chips.

Artificial intelligence is shifting from simple information processing to truly helping solve some of the most difficult reasoning problems in the world, thus benefiting all mankind.

References:

https://deepmind.google/discover/blog/gemini-achieves-gold-level-performance-at-the-international-collegiate-programming-contest-world-finals/

https://x.com/merettm/status/1968363783820353587

https://x.com/MostafaRohani/status/1968360976379703569

https://x.com/GoogleDeepMind/status/196836177632132