The leap forward of OpenAI’s O3: what does it mean for education and AI?

OpenAI’s latest language model, O3, has once again pushed the boundaries of artificial intelligence. With an impressive score of 85% on the ARC test (Abstraction and Reasoning Corpus), this model shows an unprecedented ability to adapt to new tasks. Francois Chollet, the creator of the ARC test and formerly one of OpenAI’s fiercest critics, has even changed his mind:

“O3 has shown a surprising and important leap forward in AI capabilities, with an ability to adapt to new tasks that we have never seen before in GPT models. We need to update all our assumptions about AI capabilities for O3.”

What makes O3 so special, and what are the implications for education? Let’s explore that further.

Transcending complexity: the FrontierMath test

Another impressive result comes from the FrontierMath test, which mathematician Terence Tao called ‘the hardest mathematical benchmark ever’. Where earlier AI models got no further than 2% correct, O3 managed a score of 25.2%. Terence Tao, recognised worldwide as one of the most brilliant thinkers, predicted that AI systems would struggle with this for ‘at least several years’. Yet O3 has exceeded these expectations.

This raises questions for education: can models such as O3 support complex problem-solving? Can we use AI to strengthen students’ mathematical insight through guidance and adaptive feedback?

Matching human programming performance

Another astonishing result: OpenAI’s O3 scored 2727 on Codeforces, a platform for competitive programming. This result is comparable to the performance of the 175th best human programmer in the world. This level of complex problem-solving offers opportunities for education in programming and computational thinking.

Think of applications such as:

• Peer tutoring with AI: O3 can give programming feedback at an advanced level.
• Differentiation: Personalised guidance based on the student’s level.

O3 available within Azure in 2025

From 2025, OpenAI’s O3 model will be available within Microsoft Azure. This platform is a better fit with the EU AI Act and the GDPR, making it easier for educational institutions to use AI responsibly and securely. For more information on setting up Microsoft Azure and connecting GPT-4o, watch my knowledge clip:

Reflection for education

These breakthroughs force us to rethink the role of AI in education. How can we use AI as a partner in learning, not as a replacement for human creativity and critical thinking? It is clear that we can embrace AI not only as a tool, but also as a didactic innovation.

Let’s explore these technologies critically but with an open mind, always with a view to human-centred learning. O3 has proven that AI is capable of unprecedented performance, but the human learning process remains central.

---

Sources (APA style):

• Chollet, F. (2024). Personal statement on OpenAI’s O3.
• Tao, T. (2024). FrontierMath test results. Retrieved 5 January 2025.
• OpenAI. (2024). Codeforces results of O3. Retrieved 5 January 2025.