IndiaAI Foundation Model Pillar Empanels Three More Indigenous Models

Startup Stories Jun 5, 2025 Add to Reading List

New Delhi, 5 June 2025 – The IndiaAI Foundation Model pillar, a key initiative under the Government of India’s IndiaAI Mission, is steadily building a sovereign AI ecosystem by developing and deploying foundational models trained on India-specific data. Since the launch of the Call for Proposals, 506 applications were received by 30 April 2025. On 26 April 2025, Sarvam AI was selected to build India’s Sovereign LLM Ecosystem—an open-source, 120 billion-parameter model designed to enhance governance and public service access through flagship use cases such as “2047: Citizen Connect” and “AI4Pragati.” Sarvam AI’s earlier releases, Sarvam-1 (2 billion parameters) and Sarvam-M (24 billion parameters) with hybrid reasoning capabilities, laid the groundwork for this next-generation model.

Following a rigorous multi-stage expert evaluation, the Foundation Model Committee has now selected three additional indigenous proposals:

Soket AI

Model Size: 120 billion parameters
Objective: Develop India’s first open-source foundation model optimized for the nation’s rich linguistic diversity.
Target Sectors: Defense, healthcare, and education, with specialized sub-models for multilingual text understanding and domain-specific knowledge retrieval.

Gnani AI

Model Size: 14 billion parameters
Objective: Build a Voice AI foundation model delivering real-time, multilingual speech processing with advanced reasoning.
Key Features: Real-time speech-to-text, text-to-speech, and intent detection across 15 major Indian languages—enabling conversational AI assistants for government helplines, telemedicine platforms, and accessibility tools.

Gan AI

Model Size: 70 billion parameters
Objective: Create a multilingual foundation model powering “Superhuman TTS (Text-to-Speech)” capabilities that surpass current global leaders.
Aspirations: Achieve state-of-the-art naturalness and intelligibility, with advanced prosody, emotional expressiveness, and seamless code-switching support for Indian languages.

“By selecting these three models alongside Sarvam AI, IndiaAI is advancing its mission to establish a truly sovereign AI framework—one that reflects our nation’s linguistic plurality and addresses critical public-sector priorities “ India is significantly ramping up its compute infrastructure under the IndiaAI Mission, with the addition of 15,000 new GPUs, taking total capacity to nearly 34,000 GPUs, Minister for Electronics and IT Ashwini Vaishnaw said on Wednesday. Seven firms, including Sify Digital Services and Yotta Data Services, have been selected in the second round of empanelment to supply GPUs sourced from Nvidia,

With these selections, the IndiaAI Foundation Model pillar now counts four empaneled partners:

Sarvam AI (120B parameters)
Soket AI (120B parameters)
Gnani AI (14B parameters)
Gan AI (70B parameters)

**Two additional entities—**Thore Network AI and Wadhwani AI—are next in line for empanelment, pending successful completion of technical reviews and compliance conditions outlined in their respective Requests for Empanelment (RFEs). “We expect more startups to join the Foundation Model pillar in the coming weeks, further strengthening India’s AI ecosystem,”

About the IndiaAI Foundation Model Pillar
Under the IndiaAI Mission, the Foundation Model pillar seeks to:

Develop open-source, India-centric foundational models across multiple modalities (text, speech, vision).
Foster collaboration between academia and industry to build transparent, ethical, and verifiable AI systems.
Democratize access by making models, datasets, and evaluation tools available to government ministries, startups, and educational institutions.

Selected partners receive grant funding, GPU allocation, cloud credits, and access to IndiaAI’s evaluation infrastructure. In return, they commit to open-source licensing, periodic model audits, and delivering high-impact use cases—such as improved governance, public health solutions, and next-generation education tools.

Looking Ahead: IndiaAI’s Own Model Launch
The IndiaAI Foundation Model Committee is also developing its own flagship model, slated for release between November–December 2025. This IndiaAI-branded foundation model will be trained end-to-end on pan-Indian datasets and optimized for critical use cases spanning public administration, citizen services, and digital inclusion.

Next Steps & Timeline

September 2025: Alpha releases published for community testing and feedback.
November–December 2025: Launch of IndiaAI’s own foundation model on the IndiaAI Model Zoo.

For more information on the IndiaAI Mission or to explore collaboration opportunities, please visit https://indiaai.gov.in