Aryabhata 1.0: The AI Revolutionizing JEE Math Prep

Aryabhata 1.0: A Leap Forward in JEE Math Preparation

PhysicsWallah (PW) has launched Aryabhata 1.0, a compact language model designed to assist students preparing for the JEE Main Mathematics exam. This model, named after the renowned ancient Indian mathematician, is a significant step in PW’s mission to create Small Language Models (SLMs) tailored to the specific needs of Indian learners. With a 90.2% accuracy rate in the April 2025 session, Aryabhata 1.0 is setting new benchmarks in AI-driven education.

The Development of Aryabhata 1.0

Aryabhata 1.0's development is rooted in a deep understanding of the Indian education system and the challenges faced by students preparing for competitive exams. The model was trained on a single H100 GPU and uses 130,000 curated question-answer pairs, leveraging a combination of supervised fine-tuning (SFT), reinforcement learning (RL), and rejection sampling. These techniques ensure that the model’s reasoning aligns with pedagogical patterns, making it highly effective in a real-world educational context.

Key Features and Performance

• High Accuracy**: Aryabhata 1.0 achieved 86% accuracy in the January 2025 session and improved to 90.2% in the April session. This performance is a testament to the model’s robust training and fine-tuning processes.
• Pedagogical Alignment**: The model’s training data and techniques are designed to align with how students are taught for competitive exams, ensuring that the AI’s reasoning is grounded in practical, real-world scenarios.
• Accessible Technology**: With only 7 billion parameters, Aryabhata 1.0 is a compact model that can be trained and deployed efficiently, making advanced AI technology accessible to a broader range of students.

Expanding the Horizon

PhysicsWallah’s vision for Aryabhata 1.0 extends beyond JEE Main. The company is working to expand the model to cover JEE Advanced and a broader range of mathematical domains. This expansion aims to provide comprehensive support for students across various competitive exams, further democratizing access to high-quality educational resources.

Community Engagement

PW is inviting educators, developers, and researchers to test Aryabhata 1.0 and provide feedback. This collaborative approach ensures that the model continues to evolve and improve, addressing the diverse needs of Indian learners. The goal is to build a suite of subject-centric SLMs that can support learning across multiple disciplines in an AI-driven education system.

The Bottom Line

Aryabhata 1.0 represents a significant milestone in the integration of AI into the Indian education system. By providing high-accuracy, pedagogically aligned support, the model is helping students excel in competitive exams and paving the way for a more inclusive and effective learning environment.