Indian startups Soket Labs and ZenteiQ launch domestic AI models
Soket Labs and ZenteiQ have released new large language models developed under the Indian government’s IndiaAI Mission.
Startups Soket Labs and ZenteiQ AiTech Innovations have unveiled new domestically developed large language models (LLMs) in New Delhi. The release is a component of the Indian government’s IndiaAI Mission, an initiative designed to foster local foundation models and reduce reliance on foreign technology.
Soket Labs’ EKA-1 model, featuring 26.3 billion parameters and trained on 7.5 trillion tokens, supports English and all 22 scheduled Indian languages. CEO Abhishek Upperwal stated that the model is intended for use in mathematics, coding, and reasoning, specifically targeting the requirements of India’s banking and cybersecurity sectors.
ZenteiQ’s models, Dhīr 2B and Dhīr 5B, utilize a lightweight architecture designed for on-device deployment. CEO Sashikumar noted that these models operate across various computing platforms without requiring cloud connectivity, an approach intended to improve privacy and accessibility for government and education applications.
While the government projects the Indian enterprise AI market to reach $71 billion by 2030, the companies did not provide benchmark results or pricing details at the launch. The announcement coincides with increased activity in the local AI space, including recent releases from companies like Sarvam AI, as businesses seek more control over auditability and output in enterprise workflows.