India Launches Indigenous On-Device AI Models EKA-1, Dheer 2B & 5B for Smartphones
In a historic milestone advancing India’s technological self-reliance and sovereign computing framework, the Ministry of Electronics and Information Technology (MeitY) has officially launched three indigenous artificial intelligence models capable of executing high-complexity tasks directly on mobile devices without requiring active mobile data or Wi-Fi connectivity. Developed by domestic deep-tech innovators under the aegis of the IndiaAI Mission, the newly debuted suite—comprising EKA-1, Dheer 2B, and Dheer 5B—is engineered to tackle advanced mathematical reasoning, multilingual programming, scientific computing, and cybersecurity analysis entirely on-device. Commending the development cohorts, MeitY Secretary S. Krishnan emphasized that deploying localized offline models directly fortifies India's strategic digital autonomy while tailoring AI utilities to the country's diverse digital and linguistic architecture, confirming that all three models have been made publicly accessible to startups, developers, students, and research bodies via the unified AIKosh platform.
EKA-1: A 26-Billion Parameter Powerhouse Trained on 7.5 Trillion Tokens Across 100+ Languages
Setting a new national benchmark in high-scale foundation modeling, EKA-1 emerges as an expansive 26-billion parameter architecture developed by Bengaluru-based AI enterprise Soket AI (Sackett). Trained on an extraordinary corpus of 7.5 trillion tokens, the model fluently navigates over 100 Indian regional dialects, international languages, and complex computer programming syntaxes, performing heavy mathematical computation and logical coding while retaining full functionality without cloud connectivity. Outlining the model's architectural focus, Soket CEO Abhishek Uppal noted that EKA-1 is purpose-built to address sensitive operational workflows across Indian banking, financial services (BFSI), and enterprise cybersecurity ecosystems, ensuring end-to-end data privacy by keeping enterprise data processing strictly within local device parameters.
Dheer 2B and Dheer 5B: Lightweight Mobile-First Intelligence Built for Zero-Connectivity Zones
Complementing heavy foundation computing, home-grown innovator GenteIQ has deployed Dheer 2B and Dheer 5B, two ultra-efficient on-device models optimized to run on standard consumer smartphone hardware. Built and fine-tuned using Google Cloud's advanced Tensor Processing Units (TPUs), the Dheer series supports English and 10 major Indian vernacular languages, specializing in rapid document comprehension, localized information retrieval, and prompt question-answering. GenteIQ CEO Prof. Sasikumar highlighted that the lightweight footprint of the Dheer models eliminates external cloud reliance, bridging critical digital divides by delivering instant, dependable generative computing to remote rural geographies and areas with constrained internet bandwidth.

