Marking a notable step forward in building domestic technological capabilities, the IndiaAI Mission operating under the Ministry of Electronics and Information Technology (MeitY) officially introduced three homegrown artificial intelligence foundation models on Tuesday. The newly launched models include EKA-1, Dhīr 2B, and Dhīr 5B. The initiative aligns with the national priority of cultivating localized artificial intelligence capacity and reducing reliance on foreign technological frameworks. IT Secretary S Krishnan attended the official launch event alongside senior ministry officials, representatives from IndiaAI, and executive leadership from the developer firms.
IndiaAI Mission and Public Compute Infrastructure
The creation of these systems was facilitated through the IndiaAI Innovation Centre Foundation Models pillar, a strategic programme dedicated to developing indigenous artificial intelligence and large language models. Officials highlighted that the deployment of subsidized national AI compute resources combined with targeted research support played an instrumental role in bringing the models to completion. The government emphasized that shared public AI infrastructure serves as an essential catalyst for accelerating domestic technological innovation.
IT Secretary S Krishnan stated during the proceedings that such indigenous models reinforce India's technological and strategic autonomy in the artificial intelligence sector. He noted that the availability of domestic models creates significant opportunities to enhance productivity and operating efficiency across government administration, industrial sectors, academic research, and formal education. Krishnan also highlighted that platforms like AIKosh are positioned to widen ecosystem access, enabling researchers, university students, technical innovators, and the public to test these models and build specialized applications upon them.
Soket Labs Develops EKA-1 for Indian Enterprise Demands
Selected under the foundation models pillar, Soket Labs Technology and Research engineered the EKA-1 model specifically tailored to address Indian operational environments. The architecture features 26.3 billion parameters and underwent pre-training utilizing approximately 7.5 trillion tokens. In terms of linguistic range, EKA-1 provides full support for English and all 22 scheduled Indian languages recognized by the Constitution. Soket categorizes the system as an open, sparse hybrid language model optimized for mathematical processing, coding tasks, complex reasoning, and autonomous agentic workflows.
A key operational advantage of EKA-1 is its capability to execute inferencing directly on local edge hardware, enabling specific functions to run reliably in the total absence of mobile cellular data or Wi-Fi connections. Soket Chief Executive Officer Abhishek Upperwal stated that the model directly addresses the rigorous functional demands of India's banking, financial services, and cybersecurity industries. Abhishek Upperwal explained that the model aligns with targeted sector needs and mirrors the requirements of India's evolving digital landscape.
ZenteiQ Releases Dhīr 2B and 5B for Secure On-Premise Use
In parallel, ZenteiQ AiTech Innovations launched Dhīr 2B and Dhīr 5B as open-source foundation models designed to support English alongside 10 Indian languages. Developed as secure, on-premise multilingual digital assistants, both variants focus on document-grounded tasks and offline workflow automation. Their core utility centers on organizational knowledge retrieval, administrative document processing, and facilitating enterprise or citizen service delivery mechanisms.
According to official project details, the primary design focus was ensuring dependable performance in low-bandwidth, disconnected, or security-sensitive operational settings, with practical deployments planned across defense agencies, public administrative bodies, industrial enterprises, and academic institutions. ZenteiQ Chief Executive Officer Prof. Sashikumar noted that the models operate effectively on both GPUs and TPUs. Prof. Sashikumar pointed out that their lightweight architecture permits direct installation on smartphones, eliminating the necessity for continuous cloud server connectivity.
Local Data Privacy and Broadened Accessibility
Government officials noted that running computational workloads directly on local hardware ensures data remains strictly confined to the host device, providing robust guarantees for personal privacy and institutional security. Furthermore, localized offline execution bridges the digital divide for institutions operating in remote geographies with weak telecommunications infrastructure. Linking these releases to wider public digital initiatives, officials reaffirmed that distribution via AIKosh will expand developer adoption across the country.
