Chinese tech giant ByteDance released a technical report for its Seed 1.6 series of AI models last week. In this report, the company stated the model’s performance across various evaluations, which includes how it scored on various college entrance tests.
The Seed 1.6-Thinking (reasoning) along with Google’s Gemini 2.5 Pro, Anthropic’s Claude Opus 4, and OpenAI’s o4-mini-high models were tested on the Indian Institute of Technology’s Joint Entrance Examination (IIT-JEE) 2025 Advanced examination.
It turns out, Gemini 2.5 Pro scored 336.2 out of 360 marks, outperforming the student who topped the exam with 332 marks in 2025.
The Seed 1.6 Thinking model scored 329.6 marks and stood in second place, ahead of Claude Opus 4 (314.4) and o4-mini-high (308.4).
ByteDance tested the models with image inputs and evaluated their multimodal and reasoning abilities.
While all the objective questions on which the model was tested were sampled five times, the report shared the average score for each sample. Moreover, the marks were assigned according to standard JEE exam rules, which involve deducting marks for incorrect answers.
The IIT-JEE Advanced examination is widely regarded as one of the most difficult in the world. This is the second stage of the entrance examination for IITs in India.
The top 2.5 lakh candidates appearing for the first stage of the exam, the IIT-JEE Mains exam, are qualified for the advanced stage. Applicants are tested across physics, chemistry, and mathematics. Acceptance rates inside the IITs are generally less than 2%.
The Gemini 2.5 Pro model also scored the highest marks in chemistry, whereas the Seed-1.6-Thinking scored the highest in maths. The latter model also ranked fourth overall when compared to students who appeared for the exam.
The post The Topper of IIT-JEE Advanced is…Gemini 2.5 Pro! appeared first on Analytics India Magazine.
