India releases three new homegrown foundation models

A push for technological autonomy

The IndiaAI Mission, operating under the Ministry of Electronics and IT (MeitY), recently unveiled three new indigenous foundation models. These models represent a significant step in the country's effort to build its own artificial intelligence capabilities rather than relying solely on external providers. The release was marked by an event attended by high-ranking ministry officials and the leadership of the developing companies.

The initiative is part of the IndiaAI Innovation Centre (Foundation Models) pillar. This specific program is designed to foster the creation of large language models and other foundational AI technologies within India. By supporting local development, the government aims to strengthen the nation's strategic autonomy in the global AI landscape.

The models were developed by two distinct companies selected for the program. Soket Labs Technology and Research produced the EKA-1 model, while ZenteiQ AiTech Innovations developed the Dhīr 2B and Dhīr 5B models. Each company focused on different use cases, ranging from financial services to secure government automation.

EKA-1 and linguistic diversity

EKA-1 is a large-scale model featuring 26.3 billion parameters and was trained on approximately 7.5 trillion tokens. One of its primary strengths is its linguistic reach, as it covers all 22 scheduled Indian languages in addition to English. This broad support makes it a versatile tool for a diverse population.

Soket Labs describes EKA-1 as an open, sparse hybrid language model. This means it is designed to handle complex tasks such as math, coding, reasoning, and agentic use—where the AI can perform actions to achieve a goal. It is specifically tailored to understand the evolving digital ecosystem of the country.

The developers have specifically aimed the model at the banking, financial services, and cybersecurity sectors. According to Soket CEO Abhishek Upperwal, the model addresses specific requirements within these industries. It is built to handle the unique nuances of India's financial landscape while maintaining high performance across different types of reasoning tasks.

Offline capabilities and privacy

The Dhīr 2B and Dhīr 5B models from ZenteiQ focus on a different set of priorities: security and offline functionality. These models are designed to enable automation, knowledge retrieval, and various services for citizens and enterprises in sensitive environments. This includes use cases in defense, government, industrial, and educational sectors.

A key technical feature of these models is their ability to run without a cloud dependency. They can be deployed as secure, on-premise multilingual assistants that support English and 10 Indian languages. Because they do not require a constant connection to a remote server, they are ideal for areas with low connectivity or for handling highly classified information.

ZenteiQ CEO Prof. Sashikumar noted that these models can run on various computing platforms, including GPUs and TPUs. Their lightweight architecture allows for on-device deployment on mobile phones. This setup ensures that users can access AI locally, which keeps data on the device and enhances overall privacy and security for the end user.

Impact on infrastructure and users

All three models released under the IndiaAI Mission support on-device deployment. This capability is significant because it improves accessibility for people who may not have reliable mobile data or WiFi connections. By performing certain inferencing—the process of generating an output from an input—on local hardware like smartphones, the models become more useful in rural or remote settings.

For government and industrial users, the move toward indigenous models means they can maintain tighter control over their data. Instead of sending sensitive information to a third-party cloud, organizations can host the AI on their own hardware. This reduces the risk of data leaks and ensures that the technology aligns with national security standards.

IT Secretary S Krishnan stated that these models strengthen India's technological and strategic autonomy. The goal is to create opportunities to improve productivity and efficiency across various sectors. By providing tools that work locally and in multiple languages, the mission seeks to democratize AI access while protecting the nation's digital interests.

The path toward local AI

The release of EKA-1 and the Dhīr series marks a transition toward a more self-reliant AI infrastructure. By focusing on

html_content_note_for_user_to_see_this_is_the_final_section_the_source_is_limited_so_this_summarizes_the_implications_based_on_the_text_provided_the_models_provide_a foundation for both commercial and public sector growth. The participation of private firms like Soket Labs and ZenteiQ shows a collaborative approach between the government and the private sector to build these tools.

Moving forward, the success of these models will likely depend on how well they perform in real-world applications like banking and defense. The ability to handle 22 languages and provide offline automation gives India a unique position in the global AI market. It allows for the creation of specialized tools that are culturally and linguistically relevant to the local population.

As these models become more integrated into the economy, they may set the standard for how other nations approach indigenous AI. The focus on privacy, local deployment, and linguistic diversity provides a blueprint for building AI that serves specific national needs. The IndiaAI Mission will continue to oversee the development of these capabilities to ensure they meet the country's evolving digital requirements.