editorially independent. We may make money when you click on links
to our partners.
Learn More
OpenAI and Google DeepMind stood out at the International Collegiate Programming Contest world finals in Baku, Azerbaijan, last week, with OpenAI’s models solving all 12 of the problems at the world’s top university coding competition.
In attendance was OpenAI Research Lead Ahmed El-Kishky, who spoke to our colleagues at The Neuron in a podcast interview about the win and what OpenAI learned from the competition.
OpenAI’s preparation for the contest
Generative AI has made significant progress in competitive math and programming over the last year. Earlier versions, such as GPT-4, did so poorly that its attempts would crash the sandbox computer used in the competitions, El-Kishky said.
This year, “people were nervous” on the small OpenAI team that traveled to Azerbaijan, El-Kishky noted.
Still, circumstances were very different from those in 2024. The researchers had pre-trained the models they would use for the competition, including those based on reinforcement learning.
A week before the event, the team did a dry run, “spending long nights trying to make sure everything’s working,” El-Kishky said.
Two models tackled the problems: GPT-5 and an experimental reasoning model that is not publicly available.
Testing GPT-5 against a new reasoning model
“In this case, we knew that GPT-5 can go a long way, but we wanted to not just limit it there,” said El-Kishky. “We wanted to also test how well reasoning models perform. So we tried out both.”
Both models attempted to solve the problem multiple times. The experimental reasoning model decided which solution to submit. GPT-5 was faster, but the experimental model was more powerful. GPT-5 couldn’t complete the last problem, but the experimental model did on its second attempt.
The competition was an interesting example of problems that are difficult for humans but not so difficult for an AI — and vice versa, El-Kishky said.
It also showed how far AI reasoning capabilities have come. Reasoning models and models with reinforcement learning work better because of their similarity to human problem-solving, El-Kishky explained.
“When you have a difficult problem, no human just immediately gives an answer,” he said. “If I asked you to multiply a four-digit number by a five-digit number, you wouldn’t give me an answer instantly.”
The breakthrough was reinforcement learning, or the ability to “mimic how a human thinks through these problems. They try things out, and maybe they go down wrong directions to dead ends, they course correct,” El-Kishky said.
Programming competitions are a chance to benchmark AI progress
True to OpenAI’s mission statement, El-Kisky said the real purpose of AI models is to increase human knowledge, not win programming contests.
“The competitions are not really an end goal in themselves,” El-Kishky said. “They’re just sort of a way to benchmark progress.”
The experimental model used in the competition won’t be released to the public, but the research behind it may one day be applied to other models and, eventually, to ChatGPT.
Teaching models to code autonomously is an important research topic for the company as a whole, El-Kishky noted. OpenAI aims to develop AI agents that can operate independently for days, weeks, or months.
More from El-Kishky
Check out The Neuron’s full interview with El-Kishky, which includes what it’s like to work at OpenAI and details about other competitions employees have participated in.
Spotlight on another episode of The Neuron podcast: Microsoft’s AI chief Mustafa Suleyman outlines his vision for the future of artificial intelligence and the responsibilities that come with it. His remarks could shape industry expectations on innovation and governance.