The Compact Language Model Focuses on Competitive Exams
As an artificial intelligence system built by Google DeepMind makes headlines for scoring a “gold medal” at the International Mathematical Olympiad 2025, India’s PhysicsWallah is quietly building its own mathematical AI arsenal. The edtech firm has launched Aryabhata 1.0, a compact language model designed to assist students preparing for JEE Main Mathematics.
This launch marks the platform’s first step in integrating AI-powered learning for competitive exams through small, specialized models.
Built with India’s Students in Mind
Unlike the massive foundation models from Big Tech, Aryabhata 1.0 is deliberately compact. With 7 billion parameters, it runs on a single Nvidia H100 GPU, making it lightweight yet effective.
The model’s focus? JEE Main Maths — one of India’s most competitive entrance exam sections. It was trained on 130,000 curated question-answer pairs, a dataset shaped specifically for Indian exam patterns.
Despite its smaller size, Aryabhata 1.0 scored 86% in the January 2025 session and 90.2% in the April session — putting it close to top-tier student scores.
From Aryabhata’s Legacy to AI-Powered Classrooms
Aryabhata 1.0 draws its name from India’s legendary mathematician, aligning with PhysicsWallah’s vision to localize AI for real-world learning.
“Sometimes, the right support at the right moment can change a student’s entire path,” said Prateek Maheshwari, co-founder of PW. The company believes the future of AI lies in purpose-driven models designed around actual classroom challenges.
A Pedagogical Approach to AI
PhysicsWallah has taken a different route from mainstream models. Aryabhata 1.0 is trained using supervised fine-tuning, reinforcement learning, and rejection sampling — methods that ground the model’s logic in student-centric reasoning.
This approach helps Aryabhata mimic teaching patterns rather than simply spitting out answers. The aim is to support learning, not just automate it.
Beyond JEE Main: What’s Next?
The current version is just the beginning. PW plans to expand Aryabhata 1.0 to cover JEE Advanced and other mathematical domains. It also signals a broader vision — creating a suite of subject-specific Small Language Models (SLMs).
PhysicsWallah is actively inviting educators, researchers, and developers to test Aryabhata and share feedback. This open model will help shape the next version to better match evolving student needs.
A New Chapter for India’s AI in Education
The launch of Aryabhata 1.0 underscores India’s growing AI capabilities in education. While tech giants chase universal intelligence, PW is proving that smaller, focused models can deliver big results — especially when aligned with curriculum-based learning.
Aryabhata’s debut could be a turning point, where AI doesn’t replace the teacher, but becomes the tutor.
