The Indian Ministry of Electronics and Information Technology has introduced three domestically developed AI models aimed at reducing foreign technology reliance and enhancing AI accessibility in areas with poor internet connectivity as well as securing critical national sectors.
- Dhīr 2B and Dhīr 5B enable offline AI use on mobile phones.
- EKA-1 designed for secure, high-parameter tasks in critical sectors.
- Models aim to reduce dependency on foreign AI technologies.
What happened
The Government of India, through the Ministry of Electronics and Information Technology (MeitY), has launched three indigenous AI models as part of the IndiaAI Mission's Foundation Model initiative. These models—EKA-1, Dhīr 2B, and Dhīr 5B—were developed by Bengaluru-based AI firms Socket AI and ZenteiQ to serve India's unique technological needs while emphasizing data security and autonomy.
ZenteiQ’s Dhīr models are lightweight AI systems optimized to operate offline directly on mobile devices, enhancing access to AI capabilities in regions with limited internet connectivity. Meanwhile, Socket AI’s EKA-1 is a larger, 26-billion-parameter model trained on extensive data suitable for complex tasks in mathematics, coding, and reasoning and designed to be deployed in private and air-gapped environments.
Why it matters
These models represent a strategic move by India to reduce dependence on foreign AI providers, especially important in sensitive sectors such as defence, cybersecurity, and banking where data privacy and operational security are paramount. Indigenous development provides India with greater control over critical AI infrastructure and ensures alignment with national security priorities.
Furthermore, the mobile-compatible models address a significant gap for AI accessibility across India’s diverse regions, particularly in places where internet connectivity is unreliable or unavailable. This has the potential to democratize AI usage, supporting applications in day-to-day mobile interactions without requiring constant online access.
What to watch next
The EKA-1 model weights are expected to be released publicly soon, initially via a waitlist, enabling further research and adoption within secure enterprise environments. Meanwhile, ZenteiQ plans to launch a 9-billion-parameter model in December for specialized research applications, expanding the ecosystem of indigenous AI tools available in India.
Monitoring uptake across defence, banking, and cybersecurity sectors will be key to assessing impact, alongside how these technologies integrate with existing infrastructure. The ongoing development and deployment of these AI models could pave the way for further innovations and help India establish itself as a significant player in the global AI landscape while maintaining strategic digital sovereignty.