Artificial Intelligence continues to redefine boundaries — and OpenAI’s latest model, ChatGPT o3, has made headlines by securing an equivalent of All India Rank 4 in a simulated JEE Advanced 2025 test, one of the toughest engineering entrance exams in India.
The experiment, led by IIT Kharagpur graduate Anushka Aashvi, aimed to evaluate how well a powerful AI model could perform under realistic exam conditions. ChatGPT o3 was instructed to act like a genuine JEE aspirant, solving every question independently in a new chat session — no internet access, tools, or hints were allowed.
The result? A jaw-dropping 327 out of 360 marks, with perfect scores in both Chemistry and Mathematics in later phases. Even with challenging restrictions in place, the AI performed with precision and consistency, leaving many in the academic and tech communities stunned.
While this test showcases AI’s growing ability to tackle high-stakes, human-centric tasks, it also opens up serious conversations about its future role in education, assessments, and intellectual benchmarks.
On the flip side, a recent paper from Apple AI researchers titled “The Illusion of Thinking” cautions that leading AI models — including ChatGPT o3 — may not possess genuine reasoning skills. Despite appearing intelligent, these systems often falter when facing deep or ambiguous problems.
As AI tools become smarter and more accessible, the challenge for educators and learners is not just to adopt them, but to understand their limits — and use them ethically and effectively.