New Delhi: The Government of India has identified 20 indigenous sovereign AI model proposals for support under the IndiaAI Mission, as it accelerates efforts to build domestic AI capabilities alongside safeguards for responsible development and deployment.
The selected proposals include 12 large language models and 8 small language models. Among the initiatives highlighted by the government are Sarvam AI’s 30-billion- and 105-billion-parameter models, Gnani.AI’s speech-to-speech model, BharatGen’s multilingual foundation models and Avataar AI’s video-generation model. Domain-specific models for healthcare, Indian languages and agentic AI applications are also at different stages of development and deployment.
Approved in March 2024 with an outlay of ₹10,371 crore over five years, the IndiaAI Mission is built around seven pillars: compute capacity, foundation models, AIKosh, application development, FutureSkills, startup financing, and Safe and Trusted AI.
Under its compute initiative, the government has empanelled 15 service providers and supported 237 projects. A total of 93 lakh GPU hours has been sanctioned to improve access to the high-performance computing infrastructure required for developing and deploying AI systems.
The mission has also launched 12 national-level hackathons and innovation challenges, supporting the development of 62 AI prototypes and the deployment of 20 AI solutions. In talent development, 686 fellowships have been awarded across undergraduate, postgraduate and doctoral programmes at 178 institutions. More than 26 lakh people have completed the YUVA AI for All programme.
Institutional capacity is being expanded through 58 AI Centres of Excellence across states and Union Territories, in collaboration with governments and industry partners. The government has established 27 India Data and AI Labs, which have trained over 2,500 students, while work is underway on another 188 labs.
A parallel focus is being placed on managing the risks emerging from AI adoption. Thirteen Responsible AI projects have been approved at educational institutions to address bias mitigation, explainability, privacy-preserving AI, machine unlearning, deepfake detection and AI risk assessment.
These include Saakshya, a multi-agent deepfake-detection framework developed by IIT Jodhpur and IIT Madras; AI Vishleshak, designed to improve audio-visual forgery detection; and IIT Kharagpur’s real-time voice deepfake detection system.
NIT Raipur is developing algorithms to reduce bias in medical imaging and clinical decision-making, while IIT Delhi, IIIT Delhi and IIT Dharwad are working on privacy-preserving machine-learning models using federated learning. IIT Jodhpur is also developing machine-unlearning techniques to remove sensitive, outdated or unwanted information from generative foundation models.
The IndiaAI Safety Institute, announced in January 2025, is expected to coordinate safety research and work with academia, startups, industry and government bodies through a hub-and-spoke model. India’s AI governance framework follows a risk-based approach covering algorithmic bias, misinformation, deepfakes and unintended societal harm.
The latest progress update was submitted by Union Minister for Electronics and Information Technology Ashwini Vaishnaw in the Rajya Sabha on July 24, 2026.
