Google Gemini 2.5 Pro beats tops IIT JEE Advanced 2025: Check score compared to all-India topper

Gemini 2.5 Pro’s performance included scores of 119.6 in Mathematics, 108.6 in Physics, and 108 in Chemistry, securing it the first rank. In comparison, the top human score of 332, achieved by Rajit Gupta from the IIT Delhi zone, placed him just behind the AI leader. 

Google Gemini Live features
Here is how you can get Gemini AI Pro for free. (Photo source: Reuters)

Is AI smarter than humans? Based on a recent experiment, it seems that most paid forms of AI chatbots have an edge over the smartest students, with Google’s Gemini taking the edge.

In a groundbreaking development, Google’s Gemini 2.5 Pro has emerged as the top performer in the 2025 IIT Joint Entrance Examination (IIT JEE) Advanced – one that is widely regarded as one of the toughest college entrance exams in India. The AI model scored an impressive 336.2 out of 360, surpassing the highest-scoring human candidate, Rajit Gupta, who achieved 332 marks.

The results, quietly released by ByteDance this week, highlight a significant milestone in AI capabilities. The IIT JEE Advanced, a two-stage examination, attracts over 1 million candidates in its initial phase, with the top 250,000 advancing to the second stage. The exam, consisting of two three-hour sections, tests proficiency in Mathematics, Physics, and Chemistry, awarding 3 marks for correct answers and deducting 1 mark for incorrect ones.

Gemini 2.5 Pro’s performance included scores of 119.6 in Mathematics, 108.6 in Physics, and 108 in Chemistry, securing it the first rank. Close behind, Seed 1.6-Thinking scored 329.6 (rank 4), Claude Opus 4 achieved 314.4 (rank 13), and OpenAI’s O4-mini-high managed 308.4 (rank 18). 

In comparison, the top human score of 332, achieved by Rajit Gupta from the IIT Delhi zone, placed him just behind the AI leader. 

The scores, averaged from five samples, highlight the advanced reasoning abilities of these large language models (LLMs). The emergence of AI outperforming humans in such a rigorous exam has sparked varied reactions. While some hail it as a testament to technological progress, others point out the disparities in preparation — AI models, trained on vast datasets with no time constraints. In contrast, a human candidate like Gupta completed the exam under intense pressure in a single attempt.

As the academic community digests this news, calls are emerging to test these models in other challenging exams, such as the UPSC, to further explore their potential.

Get live Share Market updates, Stock Market Quotes, and the latest India News and business news on Financial Express. Download the Financial Express App for the latest finance news.

This article was first uploaded on July two, twenty twenty-five, at fifty-seven minutes past six in the evening.

Photo Gallery

View All
Market Data
Market Data