India launches three homegrown AI models for local use

A push for technological autonomy

The Ministry of Electronics and Information Technology (MeitY) recently unveiled three new AI models as part of the broader IndiaAI Mission. This initiative is a strategic move by the Indian government to build domestic capabilities in artificial intelligence. By creating homegrown tools, the country aims to lower its reliance on technology developed by foreign entities.

The launch occurred during a formal event in New Delhi where the government showcased the specific capabilities of these new models. These tools are intended to address unique challenges faced by the Indian market, such as linguistic diversity and data security. The goal is to provide infrastructure that aligns with the country's specific business and security requirements.

To facilitate this development, the IndiaAI Mission provided the participating companies with essential resources. These included funding for research-related expenses and access to subsidized computing infrastructure. This support system helps local startups compete with larger global tech firms by lowering the high costs of training large models.

The capabilities of EKA-1

One of the primary models introduced is EKA-1, which was developed by Soket Labs Technology and Research. It is described as an open, sparse hybrid language model with 26.3 billion parameters. It was trained on a massive dataset consisting of approximately 7.5 trillion tokens to ensure high performance across various tasks.

EKA-1 is specifically designed to be versatile, handling complex requirements like mathematics, coding, and reasoning. It is also built to function as an AI agent, which can perform actions on behalf of a user. The model is particularly useful for developers who need a robust foundation for building specialized applications.

A key feature of EKA-1 is its extensive linguistic reach. It supports all 22 scheduled Indian languages in addition to English. This makes it a significant tool for reaching a wider audience across the country. Soket Labs noted that the model was developed with a focus on the banking, financial services, and cybersecurity sectors.

Offline and secure AI with Dhir

ZenteiQ AiTech Innovations developed the other two models in this launch, known as Dhir 2B and Dhir 5B. These are open-source models that support English and 10 different Indian languages. They are designed to be more accessible for users who may not have constant access to high-speed internet or who require high levels of privacy.

Unlike many popular AI tools that rely on cloud servers to process information, the Dhir models can run locally on individual devices. This means they can function without an active internet connection. This capability is a major advantage for users who need to process sensitive information without it leaving their physical hardware.

Because they can run offline, these models are well-suited for environments where data privacy is a top priority. They are designed to help users search through documents, retrieve specific information, and automate various repetitive tasks. This makes them practical for a wide range of industrial and organizational applications.

Impact on government and industry

The launch of these models has significant implications for several key sectors in India. MeitY Secretary S. Krishnan stated that these models could improve productivity across government, businesses, research, and education. By providing tools that understand local languages, the government can improve service delivery to a larger portion of the population.

For the private sector, the models offer a way to integrate AI into sensitive workflows. Industries such as defense and government can use the Dhir models to handle internal data securely. This prevents the risk of sensitive information being sent to external servers for processing, which is a common concern with many current AI providers.

In the financial sector, EKA-1 provides a foundation for building tools that can handle complex reasoning and coding tasks. This could help banks automate internal processes while maintaining strict security standards. The move signals a shift toward a more localized AI ecosystem that caters to the specific needs of Indian infrastructure.

The path toward local AI

This launch represents a milestone in India's broader strategy to become a leader in AI development. By supporting local companies with infrastructure and funding, the government is creating a pipeline for homegrown innovation. This approach helps ensure that the benefits of AI are distributed across various sectors of the economy.

The variety of models released shows a balanced approach to AI deployment. While EKA-1 offers a larger, more complex model for high-level reasoning and language support, the Dhir models offer lightweight, private alternatives. This variety allows different types of organizations to choose the tool that best fits their specific technical constraints and security needs.

Moving forward, the success of these models will likely depend on how quickly they can be integrated into everyday workflows. As more companies and government bodies begin to adopt these homegrown tools, India will see a clearer picture of its independent AI capabilities. The mission continues to provide the necessary framework for these technologies to evolve and expand.