India's latest artificial intelligence models have been built with a particular challenge in mind: making advanced computing useful in a country where people work across dozens of languages, internet connectivity varies widely, and sensitive information cannot always be sent to foreign cloud servers.

Three indigenous foundation models, EKA-1, Dhīr 2B and Dhīr 5B, were unveiled in New Delhi on November 3 under the government's IndiaAI Mission. Developed by Soket Labs and ZenteiQ AiTech Innovations, the systems are intended for applications ranging from financial services and cybersecurity to education, defence and public administration.

The Ministry of Electronics and Information Technology's announcement identifies EKA-1 as the largest of the three. Developed by Soket Labs, it contains 26.3 billion parameters and was trained on approximately 7.5 trillion tokens, the units of text and other encoded information processed during AI training.

EKA-1 supports all 22 languages listed in the Eighth Schedule of India's Constitution, alongside English. Its architecture combines language-processing techniques designed for mathematics, programming, reasoning and agentic applications, where AI systems can perform sequences of tasks rather than simply answer questions.

Soket Labs is particularly interested in the banking, financial services and cybersecurity sectors. Such applications require models capable of interpreting specialised terminology, handling complex instructions and working within organisations that maintain strict controls over sensitive information.

ZenteiQ has taken a different approach with Dhīr 2B and Dhīr 5B. These smaller, open-source models support English and 10 Indian languages and are designed to operate on local computing infrastructure, including devices that do not maintain a continuous internet connection.

The names indicate their approximate scale: two billion and five billion parameters. Smaller models generally require less memory and computing power, making them easier to deploy on modest hardware.

According to ZenteiQ, the Dhīr models can run across graphics processing units and Google's Tensor Processing Units, with lightweight configurations suitable for mobile devices. Their proposed applications include document retrieval, administrative assistance and information processing in government offices, educational institutions and industrial facilities.

An offline assistant could, for example, help employees search internal documents without transmitting their contents to an external server. Such arrangements can provide organisations with greater control over their information, although protecting the underlying devices and software remains essential.


Both developers received support through the IndiaAI Innovation Centre, including subsidised access to national computing infrastructure and assistance with research expenses.

India has been expanding that infrastructure under the IndiaAI Mission, which supports domestic model development, computing resources and the sharing of datasets and software tools. The programme forms part of a wider effort to establish indigenous capabilities across the country's technology sector.

The latest releases build on India's expanding AI infrastructure and foundation-model ecosystem, where government-supported computing resources are helping researchers and startups undertake projects that would otherwise require substantial private investment.

The government has not published comprehensive independent benchmark comparisons demonstrating how the three models perform against leading international systems across all their supported languages. Their real-world performance will also depend on deployment conditions, training quality and the applications built around them.

For India, the opportunity extends beyond developing another general-purpose chatbot. EKA-1 and Dhīr offer domestic developers distinct foundations for building language-aware services and locally controlled AI systems, allowing the country's growing research ecosystem to address practical requirements using technology developed at home.