Union Minister Jitendra Singh said BharatGen, India’s first sovereign AI model, will complete text models in all 22 scheduled languages this month. Launched in 2024 and led by IIT Bombay, the initiative focuses on Indian languages and societal context, with speech and vision models already available in 15 languages under the IndiaAI Mission.

India's sovereign artificial intelligence initiative BharatGen will complete development of text-based AI models across 22 constitutionally recognised Indian languages by the end of this month, Union Minister of State for Science and Technology Dr. Jitendra Singh informed the Rajya Sabha. To recall, the BharatGen initiative is spearheaded by IIT Bombay.

Responding to a starred question during Question Hour, Dr. Jitendra Singh said the BharatGen initiative represents India's first government-owned sovereign large language model designed specifically for Indian languages and societal context. Speech and vision capabilities have already been developed for 15 languages and will be expanded further, he added.

India's first sovereign AI model

BharatGen was launched in September 2024 to bring a sovereign AI model into public service delivery and boost citizen engagement by developing foundational models in language, speech, and computer vision.

The initiative is part of the broader IndiaAI Mission, which was launched in March 2024 with a budget allocation of Rs. 10,371.92 crore. Singh explained that BharatGen has been conceived as a national foundational model, with its distinguishing feature being its sovereign character combined with focus on India's linguistic and cultural diversity.

"The initiative marks the world's first government-funded multimodal large language model project," the Minister said, adding that unlike similar models elsewhere that cater largely to linguistically homogeneous societies, BharatGen addresses India's unique diversity.

Three-component architecture

Singh informed the House that BharatGen has three principal components - text, speech, and vision - along with dedicated domain applications such as agriculture, Ayurveda, and the legal system. While text models across all 22 scheduled languages are expected to be completed this month, speech and vision models are currently available in 15 languages.

He added that the programme is structured as a dynamic process, with scope to go beyond the scheduled languages to include dialects and regional variations as more data becomes available.

IIT Bombay leads consortium

The BharatGen initiative is spearheaded by IIT Bombay under the National Mission on Interdisciplinary Cyber-Physical Systems of the Department of Science and Technology, with implementation by the TIH Foundation for IoT and IoE at IIT Bombay.

On the institutional architecture supporting the initiative, Singh said the BharatGen consortium includes participation from several premier institutions including IIT Hyderabad, IIT Madras, IIT Kanpur, IIT Mandi, IIT Indore, and IIIT Hyderabad. He said this consortium-based approach reflects an integrated "whole-of-science" and "whole-of-nation" framework rather than a region-specific effort.

Innovation hub network

The Minister further said 25 technology innovation hubs have been established as part of the broader ecosystem supporting emerging technologies such as artificial intelligence, robotics, machine learning, and cybersecurity. He noted that four of these hubs are at IIT Indore, IIT Kanpur, IIT Dhanbad, and the Indian Institute of Science, Bengaluru. These hubs have been upgraded to facilitate closer co-location of industry and research, easing the pathway for technology transfer.

Addressing GPU shortage concerns

Addressing concerns related to the availability of computational resources, particularly graphics processing units,Singh acknowledged that compute capacity is central to advanced AI development. He said the IndiaAI Mission has a dedicated compute pillar to provide access to shared computational resources at subsidized rates for eligible users.

The mission has already surpassed its initial GPU target, deploying over 38,000 GPUs available to Indian startups at subsidized rates of just Rs. 65 per hour, significantly cheaper than commercial cloud rates.