Parliament Question: Proposals received for Indigenous AI Foundation Models

GOVERNMENT OF INDIA MINISTRY OF ELECTRONICS AND INFORMATION TECHNOLOGY LOK SABHA UNSTARRED QUESTION NO. 6041 TO BE ANSWERED ON: 01.04.2026 PROPOSALS RECEIVED FOR INDIGENOUS AI FOUNDATION MODELS 6041. SHRI BAIJAYANT PANDA: Will the Minister of ELECTRONICS AND INFORMATION TECHNOLOGY be pleased to state: (a) whether the IndiaAI Mission has attracted proposals for indigenous Artificial Intelligence (AI) foundation models; (b) the number of proposals received and entities selected; and (c) the manner in which AI compute and dataset initiatives are supporting startups and research institutions? ANSWER MINISTER OF STATE FOR ELECTRONICS AND INFORMATION TECHNOLOGY (SHRI JITIN PRASADA) (a) to (c): India’s AI strategy is based on the Hon’ble Prime Minister's vision to democratize the use of technology. It aims to address India centric challenges, create economic and employment opportunities for all Indians. India’s own Foundational Models under the IndiaAI Mission: Under the IndiaAI Foundation Models pillar, the Government of India is supporting Indian startups, consortia and institutions to develop large language models, multimodal models and domain-specific small language models. The mission invited applications under the Foundation Models Pillar inviting applications from startups, researchers and organisations to develop state-of-the-art foundational AI models trained on Indian datasets. Out of 506 proposals received, twelve organisations and consortia, including startups, industry players and academic institutions have been selected for developing Large and Small Language Models based on Indian datasets. The models are being developed based on Indian datasets spanning all 22 scheduled Indian languages. The selected projects cover multilingual foundational models, speech and voice models, multimodal AI, scientific models, healthcare reasoning systems, and agentic AI platforms.The resulting AI models are expected to contribute to the open-source ecosystem by making them available through AIKosh platform for other startups and researchers. This will fuel innovation across India’s startup and research community. Details of these organisations on Annexure -1. Government support to them includes access to compute, datasets, research support and ecosystem facilitation. Models developed by Sarvam AI, BharatGen, Gnani and Socket were launched during the IndiaAI Impact Summit 2026. These models represent important progress in building India’s own AI capabilities tailored to local languages and use-cases. Models developed by Sarvam AI have demonstrated relatively high accuracy in document understanding and Indic language processing. BharatGen, led by the IIT Bombay consortium, has developed Param2 AI model. Sarvam and BharatGen models are now available on AIKosh platform for the developer community to develop AI based applications based on that. Affordable Compute Capacity for AI Development Under IndiaAI Compute Capacity Pillar, a national AI compute capacity has been operationalized through empanelled AI service providers offering GPU access on cloud at subsidized rates to eligible users. This Common compute under IndiaAI Mission is available to startups, researchers, academic institutions and government organizations. ● More than 38,000 GPUs have been empanelled through 14 AI service providers to provide shared AI compute infrastructure ● Data centres located across the country including Mumbai, Navi Mumbai, Hyderabad, Bengaluru, Noida and Jamnagar are providing these services ● IndiaAI Compute Portal (https://compute.indiaai.gov.in) enables eligible users including researchers to discover and access GPU compute resources ● Access to these GPUs is provided at subsidized rates; Average rate is approximately Rs 65 per GPU per hour, except for select high-end GPUs ● Empanelled providers are delivering storage, networking, AI platforms and other supporting services necessary for AI model development and deployment The compute capacity is being further expanded by adding 20,000 GPUs under the IndiaAI Mission (currently under process). AIKosha: AIKosha is platform which provide access AI models, development tools and other resources. These belong to various fields such as health, agriculture, and education, with safeguards for data privacy.Presently, AIKosh platform more than 280 models available for the use of researchers, startups and other stakeholders. AIKosh also provides a secure API-based access, an AI Sandbox environment for model training and experimentation. It also provides AI Guardrails toolkits, which support developers in monitoring, evaluating and constraining model behaviour during development and deployment. The details and features of AIKosh can be accessed from https://aikosh.indiaai.gov.in/. *******Annexure - 1 Proposals selected to build India’s Foundation Models: 1. Sarvam AI - Developing an open source 120 billion parameter AI model to enhance governance and public service access through use cases like "2047: Citizen Connect" and "AI4Pragati". 2. Soket AI - Developing India's first open-source 120 billion parameter foundation model optimized for the country's linguistic diversity, targeting sectors such as defence, healthcare, and education. 3. Gnani AI - Building a 14 billion parameter Voice AI foundation model delivering multilingual, real-time speech processing with advanced reasoning capabilities. 4. Gan AI – Creating a 70 billion parameter multilingual foundation model targeting "Superhuman TTS(text-to-speech)" capabilities to surpass current global leaders. 5. Avatar AI – Creating specialized “AI Avatars” up to 70B parameters, optimized for Indian languages and domains such as agriculture, healthcare, and governance. 6. IIT Bombay Consortium – Bharat Gen – Developing multilingual and multimodal models ranging from 2B to 1T parameters, with an open-source approach to support applications in agriculture, finance, legal, health, and education. 7. Fractal Analytics Ltd. – Building India’s first large reasoning model of up to 70B parameters, designed for structured reasoning, STEM disciplines, and medical problem- solving. 8. Tech Mahindra Maker’s Lab – Designing an efficient 8B parameter model for Indic languages (with a focus on Hindi dialects), alongside an agentic AI platform, Orion, for government applications. 9. Zenteiq – Developing BrahmAI, a science-driven multimodal foundation model (8B–80B parameters) to advance engineering intelligence, scientific computing, and industrial innovation. 10. GenLoop – Creating small language models (2B parameters) – Yukti (Base), Varta (Instruction), and Kavach (Guard) – to support all 22 scheduled Indian languages with native reasoning and content moderation. 11. Intellihealth – Proposing a 20B parameter model for EEG signal analysis to enable early screening of neurological disorders and advance brain–computer interface research. 12. Shodh AI – Developing a 7B parameter model for material discovery, integrating AI into experimental workflows to accelerate innovation in material sciences. *******