IndiaAI Mission: New Indigenous AI Models to Work Offline on Mobiles
India unveils EKA-1, Dhīr 2B, and Dhīr 5B AI models to enable offline mobile access and secure operations in defence and banking.
In a significant leap toward digital sovereignty, the Union Ministry of Electronics and Information Technology (MeitY) has unveiled three indigenous artificial intelligence models under the IndiaAI Mission. These models—EKA-1, Dhīr 2B, and Dhīr 5B—are engineered to break the reliance on foreign AI technologies and provide secure, high-performance capabilities tailored to India's unique requirements, including the ability to function without an internet connection.
AI in Your Pocket: The Dhīr Series
Developed by Bengaluru-based ZenteiQ, the Dhīr 2B and Dhīr 5B models are designed specifically for accessibility. Unlike most modern AI that requires massive cloud servers and constant connectivity, these compact models can operate directly on mobile devices.
Professor Sashikumaar Ganesan, Founder and Chief Scientist of ZenteiQ, explained that once a user downloads the model, the configuration happens automatically, allowing the AI to reside locally on the phone. This innovation is expected to be a game-changer for users in remote areas with limited or unstable internet access.
What's next for ZenteiQ?
- Mobile Apps: A free application for both Android and iOS users is currently under development.
- AIKosh: The model weights are already accessible through the AIKosh platform.
- Future Expansion: The company plans to launch a more powerful 9-billion-parameter model in December, specifically for engineers and researchers tackling specialized problems.
EKA-1: Powering Critical Infrastructure
While the Dhīr models focus on accessibility, Socket AI's EKA-1 is built for power and security. This 26-billion-parameter model was trained on a staggering 7.5 trillion tokens, making it highly proficient in complex reasoning, coding, and mathematics.
The primary objective of EKA-1 is to safeguard India's most sensitive sectors. Socket AI CEO Abhishek Upperwal emphasized that the model is designed to eliminate the need for foreign AI in critical areas such as:
- Defence
- Cybersecurity
- Banking and Finance
Because EKA-1 can be deployed on private servers, private clouds, and "air-gapped" environments (systems completely isolated from external networks), it ensures that sensitive data never leaves the secure perimeter of the organization.
A Commitment to "Made in India"
The development of these models represents a full-stack indigenous effort. Mayank Singh, Chief Science Officer and Co-Founder of Socket AI, noted that every stage—from data preparation and training to final evaluation—was carried out entirely within India.
To encourage further innovation, Socket AI intends to make the model weights openly available within a month, though initial access will be managed via a waitlist.
Strengthening Strategic Autonomy
MeitY Secretary S. Krishnan highlighted that these indigenous models are a cornerstone of India's strategic autonomy in the AI landscape. By building domestic capabilities, India is not only enhancing its digital infrastructure but also ensuring that its data remains under national control.
This initiative is part of a broader vision to create a versatile AI ecosystem that serves everyone—from a citizen in a remote village using a mobile phone to a defence official securing national secrets.