Indian technology startups Soket Labs and ZenteiQ AiTech Innovations have unveiled new, domestically developed large language models (LLMs) as part of the government-backed IndiaAI Mission. The models, introduced in New Delhi, aim to enhance the country’s technological autonomy in artificial intelligence.
Development and Capabilities
Soket Labs released EKA-1, a 26.3 billion parameter model trained on approximately 7.5 trillion tokens. The company describes the model as an open, sparse hybrid system capable of supporting English and all 22 scheduled Indian languages. According to Soket Labs CEO Abhishek Upperwal, EKA-1 is tailored for applications in banking, financial services, and cybersecurity, with specific functionalities for mathematics, coding, and reasoning.
ZenteiQ AiTech Innovations introduced two open-source models, Dhīr 2B and Dhīr 5B. These models support English and 10 Indian languages. ZenteiQ CEO Sashikumar stated that the lightweight architecture allows for on-device deployment on mobile phones and local operation without cloud connectivity, which the company claims improves data privacy and security. These models are intended for use in government and education sectors for tasks such as document automation and knowledge retrieval.
IndiaAI Mission Support
The Ministry of Electronics and Information Technology stated that both companies were selected under the IndiaAI Innovation Centre pillar of the IndiaAI Mission. This government initiative provided subsidized compute resources and research funding to assist in the development of the foundation models. IT Secretary S. Krishnan noted that these indigenous models are expected to increase productivity across government, industry, and research sectors.
In a related development, Sarvam AI announced the launch of Prism, an open-weight decision model designed for enterprise workflows, such as customer support routing. The announcement from the government regarding the Soket and ZenteiQ models did not include benchmark results, specific deployment timelines, or pricing information.