IndiaAI Mission releases three homegrown models: EKA-1 and Dhīr 2B and 5B
EKA-1 from Soket Labs covers all 22 scheduled Indian languages; ZenteiQ's two smaller Dhīr models are open source.
- Launched 3 Nov under IndiaAI Mission
- EKA-1: 26.3B parameters, 22 Indian languages
- Dhīr 2B/5B: open source, 10 Indian languages
The launch
On 3 November the IndiaAI Mission released three indigenous foundation models. EKA-1 from Soket Labs is described as an open, sparse hybrid model with 26.3 billion parameters, trained on about 7.5 trillion tokens, and it supports all 22 scheduled Indian languages plus English. It is aimed at maths, coding, reasoning and multi-step agent tasks, with banking, financial services and cybersecurity as focus areas.
The small models
ZenteiQ's Dhīr 2B and Dhīr 5B are open-source multilingual models supporting English and 10 Indian languages. Several outlets say they are designed to run on smartphones and offline or on-premise, which matters for secure deployments. The figures come from press coverage and ministry statements, not an official technical paper.
