Move over, ChatGPT! India’s own AI has arrived—the 'Dhir' and 'EKA-1' models can run on mobile phones without an internet connection..

India has taken a significant step toward self-reliance in the field of technology. In collaboration with the 'IndiaAI Mission,' the Ministry of Electronics and Information Technology (MeitY) has unveiled three indigenous artificial intelligence (AI) models. These AI tools have been developed domestically—with a specific focus on Indian languages ​​and local business requirements—to reduce India's dependence on foreign technology.

The three models—named EKA-1, Dhir 2B, and Dhir 5B—were officially introduced at an event in New Delhi. S. Krishnan, Secretary at the Ministry of Electronics and Information Technology, along with other government officials, explained that these tools are designed to enhance productivity across government offices, research centers, schools, and the private sector. Notably, two of these models are compact enough to operate directly on a user's device without requiring an internet connection.

What are the features of the EKA-1 AI model?
Among the tools launched, EKA-1 is the largest model, developed by the Bengaluru-based startup Soket Labs. It is a highly powerful, open, and sparse hybrid language model engineered to handle complex tasks such as advanced mathematics, software programming, and logical reasoning. Furthermore, it can function as an AI agent, capable of autonomously executing a series of tasks.

EKA-1 is substantial in scale; it operates on 26.3 billion parameters and has been trained on 7.5 trillion tokens of text data. Its most significant unique selling proposition (USP), however, is its linguistic capability: the model can readily process English as well as all 22 Indian languages ​​recognized under the Eighth Schedule of the Constitution of India. Given this exceptional reasoning capability, Soket Labs has designed it with major sectors—such as banking, financial services, and cybersecurity—in mind.

How do the Dhir 2B and Dhir 5B models differ?
While EKA-1 is a massive and complex system, the Dhir 2B and Dhir 5B models are built for entirely different use cases. Developed by ZenteiQ AiTech Innovations, these open-source models are relatively compact and designed for everyday, local-level use. In addition to English, they can understand 10 different Indian languages.

Rather than focusing on heavy-duty coding or complex mathematics, the Dhir models are designed to facilitate easy searching within documents, extract relevant information, and automate mundane, repetitive daily tasks. Although their exact parameters were not disclosed at the time of launch, the numbers in their names (2B and 5B) indicate their size (in billions of parameters), making them well-suited for operation on standard consumer devices.

Offline functionality is a game-changer.
The most significant feature of the Dhir 2B and Dhir 5B models is that they do not require an active Wi-Fi connection or a large cloud server to function. You can download them directly onto your mobile phone or laptop and use them entirely offline.

When an AI operates locally on your device, your personal data never leaves that device. This technology offers robust security, which is particularly crucial for organizations that handle highly sensitive or private data. Thanks to this superior built-in privacy, the Dhir models are ideal for the defense sector, government departments, and the education sector—areas where data security is of the utmost importance. Building and training large AI models from scratch is an extremely expensive undertaking, as it requires the immense computing power of supercomputers. To help Indian startups compete with global giants, the government’s ‘IndiaAI Mission’ has stepped in by providing subsidies for computing infrastructure and funding for research-related expenses. This government support will prove to be a milestone in strengthening India’s own indigenous AI ecosystem.

Disclaimer: This content has been sourced and edited from NDTV India. While we have made modifications for clarity and presentation, the original content belongs to its respective authors and website. We do not claim ownership of the content.