What is the story about?
The Office of the Principal Scientific Adviser (PSA) to the Government of India released a white paper titled ‘Advancing Indigenous Foundation Models’ on Thursday, March 13. This document is part of an ongoing AI Policy White Paper Series aimed at shaping India's artificial intelligence ecosystem.
The white paper identifies developing indigenous foundation models as a strategic priority to ensure inclusive growth and public good while aligning with India's values, legal framework, and security interests.
Introduction
Foundation models are large AI systems trained on massive datasets such as text, images, audio, and video, which could work horizontally across domains or vertically for specific sectors. They are considered a crucial layer in modern AI development, enabling tasks such as translation, summarisation, question answering, and text classification.
India aims to use foundation models to promote inclusive growth and public welfare while adhering to the country's values, legal framework, and security concerns.
The government aims to build indigenous foundation models trained on India-relevant datasets and governed under national frameworks. This approach seeks to ensure transparency, inclusivity, and alignment with national priorities while strengthening India’s position in the global AI ecosystem.
Focus on small and sector-specific models
Alongside large language models (LLMs), the white paper highlights the importance of small language models (SLMs) with multimodal capabilities. These are specialised models designed for sector-specific tasks and are typically more economical to run and maintain. In India’s context, SLMs can be developed for domains such as agriculture, healthcare, education, and micro, small and medium enterprises (MSMEs).
This combination of LLMs, SLMs, and multimodal models aims to support linguistic inclusion, affordability, and energy efficiency while enabling innovation in several areas, such as agriculture, health, education, climate, and urban governance.
Government push for indigenous AI development
India is actively building its own foundation-model ecosystem through public and private collaboration. Currently, many AI models used in the country are developed abroad and trained on datasets that do not adequately represent India’s diversity.
Recognising this gap, the government has prioritised the development of indigenous AI systems as part of its national digital infrastructure.
The IndiaAI Mission, approved in March 2024 with an outlay of ₹10,371.92 crore over five years, is the central initiative driving this effort. The programme aims to strengthen India’s capabilities in developing and deploying AI technologies, including foundation models.
The mission's two specific pillars, the India AI computation Portal and the AI-Kosh platform, address the need for substantial data and computation resources while developing foundation models.
Innovation initiatives and model development
Under the IndiaAI Mission, the government released a Call for Proposals in January 2025, inviting startups, researchers, and entrepreneurs to develop AI foundation models tailored to India’s needs. The initiative received 506 proposals by April last year.
In the first phase, four organisations—Sarvam AI, Soket AI, Gnani AI, and Gan AI—were selected to develop multilingual text models, voice AI systems, and advanced text-to-speech technologies.
In the second phase, eight additional initiatives were approved to develop large and small language models trained on Indian datasets covering all 22 scheduled languages.
Several indigenous AI models have already been announced. Sarvam AI introduced Sarvam-105B, a language model optimised for Indic languages. Gnani.ai launched Inya VoiceOS, a voice-to-voice model designed to process audio directly and support over 15 languages.
Fractal unveiled Vaidya 2.0, a reasoning model aimed at complex medical diagnostics and STEM applications. Tech Mahindra, in collaboration with NVIDIA, launched Project Indus, a Hindi-first AI model focused on education and culturally relevant learning.
Academic initiatives are also contributing significantly. The BharatGen project led by IIT Bombay has developed several AI models, including Param-1 for text processing, Shrutam for speech recognition, Sooktam for text-to-speech, and Patram for document understanding.
Private sector contributions
Private companies are also developing smaller, efficient language models tailored to industry applications. Zoho has released its in-house Zia LLM, designed for enterprise workflows such as data extraction and summarisation. Meanwhile, CoRover.ai has launched BharatGPT, a multilingual model trained on Indian conversational data in 12 languages and released in BF16 format.
These initiatives indicate that India’s AI roadmap includes both capability-maximising large models and deployment-ready efficiency models.
Key pillars of the approach
Instead of relying on a single model, the paper outlines an ecosystem-based strategy centred on three core components:
(i) shared compute access,
(ii) India-centric data and model repositories, and
(iii) multiple model-building efforts across text, speech, multimodal, and sectoral systems.
About Previous Papers
Prior to the release of the ‘Advancing Indigenous Foundation Models’ paper on March 13, the Office of the Principal Scientific Adviser (PSA) issued two other white papers. Previous papers in the series include:
Democratising Access to AI Infrastructure (December 2025): Democratise access means that the AI infrastructure must be treated as a shared national resource, encouraging innovators to construct local-language tools, adapt assistive technology, and develop solutions that address India's diverse requirements.
Strengthening AI Governance Through Techno-Legal Framework (January 2026): The White Paper outlined India’s approach to building a trusted, accountable, and innovation-aligned artificial intelligence (AI) ecosystem.
The white paper identifies developing indigenous foundation models as a strategic priority to ensure inclusive growth and public good while aligning with India's values, legal framework, and security interests.
???????? ???????????????? ???????? ???????????? ????????-???????????????????? ???????? ???????????????????????? ???????????????????? ???????????????????? ????????????????????????,
???????????? ???????????????????????? ???????? ???????????? ???????????????????????????????????? ???????????????????????????????????????? ???????????????????????????? ???????? ???????????? ???????????????????????????????????????? ???????? ???????????????????? ???????????????????????????????? ???? ???????????????????? ???????????????????? ????????…
— Office of Principal Scientific Adviser to the GoI (@PrinSciAdvOff) March 13, 2026
Introduction
Foundation models are large AI systems trained on massive datasets such as text, images, audio, and video, which could work horizontally across domains or vertically for specific sectors. They are considered a crucial layer in modern AI development, enabling tasks such as translation, summarisation, question answering, and text classification.
India aims to use foundation models to promote inclusive growth and public welfare while adhering to the country's values, legal framework, and security concerns.
The government aims to build indigenous foundation models trained on India-relevant datasets and governed under national frameworks. This approach seeks to ensure transparency, inclusivity, and alignment with national priorities while strengthening India’s position in the global AI ecosystem.
Focus on small and sector-specific models
Alongside large language models (LLMs), the white paper highlights the importance of small language models (SLMs) with multimodal capabilities. These are specialised models designed for sector-specific tasks and are typically more economical to run and maintain. In India’s context, SLMs can be developed for domains such as agriculture, healthcare, education, and micro, small and medium enterprises (MSMEs).
This combination of LLMs, SLMs, and multimodal models aims to support linguistic inclusion, affordability, and energy efficiency while enabling innovation in several areas, such as agriculture, health, education, climate, and urban governance.
Government push for indigenous AI development
India is actively building its own foundation-model ecosystem through public and private collaboration. Currently, many AI models used in the country are developed abroad and trained on datasets that do not adequately represent India’s diversity.
Recognising this gap, the government has prioritised the development of indigenous AI systems as part of its national digital infrastructure.
The IndiaAI Mission, approved in March 2024 with an outlay of ₹10,371.92 crore over five years, is the central initiative driving this effort. The programme aims to strengthen India’s capabilities in developing and deploying AI technologies, including foundation models.
The mission's two specific pillars, the India AI computation Portal and the AI-Kosh platform, address the need for substantial data and computation resources while developing foundation models.
Innovation initiatives and model development
Under the IndiaAI Mission, the government released a Call for Proposals in January 2025, inviting startups, researchers, and entrepreneurs to develop AI foundation models tailored to India’s needs. The initiative received 506 proposals by April last year.
In the first phase, four organisations—Sarvam AI, Soket AI, Gnani AI, and Gan AI—were selected to develop multilingual text models, voice AI systems, and advanced text-to-speech technologies.
In the second phase, eight additional initiatives were approved to develop large and small language models trained on Indian datasets covering all 22 scheduled languages.
Several indigenous AI models have already been announced. Sarvam AI introduced Sarvam-105B, a language model optimised for Indic languages. Gnani.ai launched Inya VoiceOS, a voice-to-voice model designed to process audio directly and support over 15 languages.
Fractal unveiled Vaidya 2.0, a reasoning model aimed at complex medical diagnostics and STEM applications. Tech Mahindra, in collaboration with NVIDIA, launched Project Indus, a Hindi-first AI model focused on education and culturally relevant learning.
Academic initiatives are also contributing significantly. The BharatGen project led by IIT Bombay has developed several AI models, including Param-1 for text processing, Shrutam for speech recognition, Sooktam for text-to-speech, and Patram for document understanding.
Private sector contributions
Private companies are also developing smaller, efficient language models tailored to industry applications. Zoho has released its in-house Zia LLM, designed for enterprise workflows such as data extraction and summarisation. Meanwhile, CoRover.ai has launched BharatGPT, a multilingual model trained on Indian conversational data in 12 languages and released in BF16 format.
These initiatives indicate that India’s AI roadmap includes both capability-maximising large models and deployment-ready efficiency models.
Key pillars of the approach
Instead of relying on a single model, the paper outlines an ecosystem-based strategy centred on three core components:
(i) shared compute access,
(ii) India-centric data and model repositories, and
(iii) multiple model-building efforts across text, speech, multimodal, and sectoral systems.
About Previous Papers
Prior to the release of the ‘Advancing Indigenous Foundation Models’ paper on March 13, the Office of the Principal Scientific Adviser (PSA) issued two other white papers. Previous papers in the series include:
Democratising Access to AI Infrastructure (December 2025): Democratise access means that the AI infrastructure must be treated as a shared national resource, encouraging innovators to construct local-language tools, adapt assistive technology, and develop solutions that address India's diverse requirements.
???????????????????????? ???????? ???????????? ???????????????????????????????????? ???????????????????????????????????????? ???????????????????????????? ???????? ???????????? ???????????????????????????????????????? ???????? ???????????????????? ???????????? ???????????????????????????????? ???? ???????????????????? ???????????????????? ???????? ???????????????????????????????????????????????????? ???????????????????????? ???????? ???????? ????????????????????????????????????????????????????????.
With AI becoming…
— Office of Principal Scientific Adviser to the GoI (@PrinSciAdvOff) December 29, 2025
Strengthening AI Governance Through Techno-Legal Framework (January 2026): The White Paper outlined India’s approach to building a trusted, accountable, and innovation-aligned artificial intelligence (AI) ecosystem.
