Google's Gemini is IIT JEE Topper, Chinese AI Model Gets 4th Rank

Google’s artificial intelligence model, Gemini 2.5 Pro, has topped India’s Joint Entrance Examination (JEE) Advanced – widely considered one of the toughest college entrance exams in the world – outperforming more than a million human candidates and edging out other frontier AI models, according to a benchmark study released by Chinese tech giant ByteDance.

The technical report, published by the Chinese technology company late last week, evaluated the performance of several leading large language models (LLMs) on the 2025 JEE Advanced exam. Gemini 2.5 Pro scored 336.2 out of a possible 360, ranking first among all participants, including the official top human scorer who achieved 332.

"Compared with the all-India human test takers, the first place scored 332 points and the 10th place scored 317 points. Gemini-2.5-Pro ​​and Seed1.6-Thinking were able to achieve top 10 scores in India. Gemini-2.5-Pro ​​performed well in Physics/Chemistry, and Seed1.6-Thinking answered all the 5 sampled math tests correctly," ByteDance said.

The JEE Advanced, taken only by the top 250,000 students who qualify through JEE Main, consists of two three-hour papers covering Mathematics, Physics, and Chemistry, with negative marking that demands precision.

Other strong performers included Anthropic’s Claude Opus 4, which scored 314.4 and ranked 13th, and OpenAI’s o4-mini-high, which came 18th with 308.4 marks.

Seed 1.6–Thinking is an upgrade over ByteDance’s earlier model, with expanded computing power and a richer training dataset that includes math, coding, puzzles, and visual content. It uses long chain-of-thought reasoning to solve complex problems, though this occasionally leads to overthinking.

To further improve its thinking ability, Seed 1.6–Thinking uses parallel decoding, allowing it to process more tokens before giving an answer. This improves performance on tough problems without needing retraining. On the challenging Beyond AIME math test set, the model improved by 8 points. It also performed significantly better on coding tasks.

Interest in advanced reasoning AI, often referred to as "superintelligence", is growing rapidly. Meta CEO Mark Zuckerberg recently announced the formation of Meta Superintelligence Labs. This will be led by high-profile hires like Alexandr Wang (former Scale AI CEO) and Nat Friedman (former GitHub CEO).

Amid the race for AI, there has been a poaching race. Meta is trying to lure AI talent from OpenAI. This in turn, has ignited one of Silicon Valley’s fiercest hiring wars. Shengjia Zhao, Jiahui Yu, Hongyu Ren, and Shuchao Bi were among the latest to leave OpenAI for Meta, as per media reports.