Fractal Analytics seeks funding for India's first reasoning model

Fractal Analytics, a Mumbai-based AI company, has proposed to create India's first large reasoning model (LRM). The total cost of this project is estimated at Rs 118.8 crore, and Fractal is seeking Rs 76.6 crore in funding from the government under the IndiaAI Mission. The LRM series will consist of three models: a small model with 2-7 billion parameters, a medium model with 20-32 billion parameters, and a large model with 70 billion parameters. The large model will be trained with up to 1 trillion tokens. This initiative is designed to enhance India's capabilities in advanced reasoning and decision-making tasks. Fractal's founder, Srikanth Velamakanni, emphasized the need for models that can "think and reason." He noted that the focus has shifted from pre-trained models to post-training models that perform complex real-world tasks. Current large reasoning models, such as DeepSeek R1, which has 671 billion parameters, show that India needs to catch up with global leaders like the US and China. The IndiaAI Mission has already received 120 applications for building sovereign AI foundation models in its latest bidding round. With 67 proposals from the first round, the total application count is now 187. The Ministry of Electronics and Information Technology aims to approve several applications by the end of March. Fractal also plans to develop a diverse dataset from major Indian examinations, aiming to create a world-class reasoning model in fields such as STEM, medical, and programming. They believe that focusing on inference-time computing will be a more cost-effective strategy as they work towards achieving artificial general intelligence (AGI) in India.