As India prepares to host the India AI Impact Summit from 15–20 February 2026 at Bharat Mandapam, New Delhi, its AI story is shifting decisively from aspiration to action. At the centre of this transformation are 12 indigenous foundation models, selected under the IndiaAI Mission from more than 500 proposals and now powered by a national compute capacity of 38,000 GPUs.
These models will be open-sourced to promote inclusive innovation, and sector-wide digital transformation, thereby forming the backbone of India’s push for AI sovereignty,
Here’s a spotlight on the 12 Made-in-India Foundation Models:
- Avataar.ai’s AI Avatars
Avataar.ai is developing a library of “AI Avatars” (specialized models) fine-tuned for Indian languages and domains like agriculture (crop advisory), healthcare (patient chatbots), and governance (public query handling). These avatars could personalize services at scale, making AI feel truly local.
- BharatGen by IIT Bombay Consortium
Led by IIT Bombay, BharatGen is building an open-source suite of multilingual and multimodal models scaling from 2 billion to a massive one trillion parameters. Its recent 2.9B bilingual LLM, PARAM-1, outperformed global peers on key benchmarks, setting the blueprint for sovereign AI with script-aware tokenizers and 25% of training corpus dedicated to Indic data. Funded with ₹1058.5 crore and supporting agriculture, finance, legal, health, and education apps, it is India’s bid for trillion-parameter innovation
- Fractal Analytics’ Reasoning Model
Fractal is pioneering India’s first large reasoning model with up to 70 billion parameters, emphasizing structured reasoning for STEM disciplines and medical diagnostics. Their Fathom-R1-14B, post-trained for just $499, delivers state-of-the-art reasoning capabilities that rival leading proprietary models, unlocking AI for policy-making, security, and healthcare breakthroughs
- Tech Mahindra’s Indic Language Model
Tech Mahindra‘s Makers Lab is designing an efficient 8 billion-parameter model honed on Hindi dialects, paired with the agentic AI platform Orion for government workflows. With ₹1.06 crore in support, it’s streamlining e-governance, from permit approvals to disaster response, in everyday Hindi.
- ZenteiQ.ai’s BrahmAI
ZenteiQ.ai’s BrahmAI is a science-driven multimodal foundation model (8-80 billion parameters) advancing engineering intelligence, scientific computing, and industrial R&D. From accelerating aerodynamic simulations to optimizing industrial fluid flows, it’s fueling India’s deep-tech R&D renaissance with physics-informed AI precision
- Genloop’s Yukti, Varta, and Kavach
Genloop‘s trio of 2 billion-parameter small language models—Yukti (base), Varta (instruction-tuned), and Kavach (guardrail for moderation)—covers all 22 scheduled Indian languages with native reasoning. Lightweight and ethical, they’re perfect for mobile apps in education and social media safety.
- NeuroDX’s EEG Analysis Model
Under Intellihealth (NeuroDX), this 20 billion-parameter multimodal model analyzes EEG signals for early detection of neurological disorders like epilepsy and Alzheimer’s. It advances affordable brain-computer interfaces, democratizing neuroscience for rural clinics and research labs alike.
- Shodh AI’s Material Discovery Model
Shodh AI‘s 7 billion-parameter model integrates AI into lab workflows to accelerate material discovery, from sustainable batteries to advanced alloys. By automating experiments, it’s accelerating India’s green tech and defense materials ecosystem.
- Sarvam AI’s Sovereign LLM Ecosystem
Developed by Sarvam AI, this open-source 120 billion-parameter model enhances governance and public services with use cases like “2047: Citizen Connect” for citizen engagement and “AI4Pragati” for progress tracking. It’s building a robust ecosystem tailored for India’s administrative needs, ensuring AI speaks the language of public policy.
- Soket AI’s Linguistic Diversity Model
Soket AI Labs is crafting India’s first open-source 120 billion-parameter foundation model, optimized for the country’s 22+ languages. Targeting defense, healthcare, and education, it promises seamless AI integration across sectors where linguistic nuances matter most, bridging urban-rural divides.
- Gnani AI’s Voice AI Foundation Model
With 14 billion parameters, Gnani.ai‘s model excels in multilingual, real-time speech processing and advanced reasoning, featuring native “speech-to-speech” capabilities. Ideal for voice-enabled services in call centers, education, and accessibility tools, it is making conversational AI inclusive for non-English speakers across India’s dialects.
- Gan AI’s Superhuman TTS Model
Gan.ai is pushing boundaries with a 70 billion-parameter multilingual model focused on “Superhuman” text-to-speech (TTS) capabilities, aiming to outpace global leaders in naturalness and expressiveness. This will transform audiobooks, virtual assistants, and content creation for India’s audio-loving audience.
These 12 models are not prototypes; they reinforce India’s commitment to:
- AI sovereignty through strategic autonomy in core models
- Inclusive innovation across languages, sectors, and communities
- Responsible governance guided by Safe & Trusted AI principles
- Scalable impact in healthcare, agriculture, public services, R&D, and education
As the world arrives in New Delhi for the India AI Impact Summit 2026, these models will demonstrate how India is moving from AI pilots to population-scale transformation.
If you have an interesting Article / Report/case study to share, please get in touch with us at editors@roymediative.com roy@roymediative.com, 9811346846/9625243429.
