News · eGov
India has launched EKA-1, Dhir 2B and Dhir 5B under the IndiaAI Mission, with the two Dhir models able to run offline on smartphones and laptops without an internet connection.
✦ AI SUMMARY 30-second read
- Three AI models — EKA-1, Dhir 2B and Dhir 5B — were unveiled at an event in New Delhi with support from the Ministry of Electronics and Information Technology.
- EKA-1, developed by Soket Labs, carries 26.3 billion parameters and was trained on 7.5 trillion tokens, supporting English and all 22 scheduled Indian languages.
- The Dhir models, built by ZenteiQ AiTech Innovations, are open-source and can process data locally on a device without cloud connectivity, supporting English and 10 Indian languages.
Why it matters: Locally developed AI that runs offline addresses India's connectivity gaps, data security concerns and the need for Indian-language support across public and private sectors.
This summary was generated by AI. Editors may review and update it.
India has introduced three domestically developed artificial intelligence models under the IndiaAI Mission, with two of them designed to operate directly on smartphones and laptops without requiring an internet connection. The models — EKA-1, Dhir 2B and Dhir 5B — were launched at an event in New Delhi backed by the Ministry of Electronics and Information Technology.
IT Ministry Secretary S. Krishnan spoke at the launch, pointing to potential uses across government departments, educational institutions, research facilities and private businesses. The models were built by two Bengaluru-based companies: Soket Labs and ZenteiQ AiTech Innovations.
The three models address distinct use cases, from advanced reasoning and software programming to document processing and the automation of routine tasks.
What was announced
EKA-1, developed by Soket Labs, is the largest of the three models. Classified as an open, sparse hybrid language model, it has 26.3 billion parameters and was trained on 7.5 trillion tokens of data. It supports English alongside all 22 scheduled Indian languages.
The model is built for demanding tasks such as mathematical reasoning, software programming, logical analysis and multi-step AI agent operations. Its stated target sectors include banking, financial services and cybersecurity.
The Dhir 2B and Dhir 5B models, developed by ZenteiQ AiTech Innovations, are open-source and support English and 10 Indian languages. Their core capabilities cover document search, information extraction and repetitive task automation. Crucially, both can be downloaded and run entirely on compatible consumer devices, with no need for continuous cloud access.
Why offline capability matters
Running AI processing on a local device rather than routing data through external cloud servers gives organisations greater control over sensitive information. This is particularly relevant for government bodies, educational institutions and any organisation that regularly handles confidential documents.
Offline functionality also removes the dependency on a stable internet connection, which remains uneven across parts of India. For public-sector and field applications where connectivity cannot be guaranteed, this capability could expand practical AI adoption.
Role of the IndiaAI Mission
Building large AI models demands substantial computing resources, infrastructure and sustained research investment — barriers that can disadvantage smaller domestic developers competing with established international firms. The IndiaAI Mission has sought to lower these barriers by offering subsidised access to computing infrastructure and financial support for research.
The initiative is designed to build a domestic AI ecosystem and encourage technologies that reflect India's linguistic diversity and industrial needs. The three models represent different points on that spectrum, with EKA-1 targeting enterprise and analytical workloads and the Dhir models prioritising everyday offline productivity.
What happens next
With all three models now available, government departments, businesses and researchers can begin evaluating them for their specific requirements. The open-source nature of the Dhir models means developers can also adapt them for a wider range of applications.
Elets will continue to follow developments around the IndiaAI Mission and the deployment of these models across sectors.