Indian Army Builds Its Own Multimodal AI With Sarvam 105B

The Defence Services Staff College (DSSC), a military training institution in Wellington, Tamil Nadu, has launched DRONA 2.0, a multimodal generative AI system integrating Sarvam AI’s 105B large language model with open-source models for defence education and research.

DRONA 2.0 was launched by Lt Gen Manish Erry, Commandant of DSSC, in the presence of Sarvam AI co-founder Vivek Raghavan. DSSC and Sarvam AI also signed a memorandum of understanding to collaborate on further development of DRONA GPT and other AI applications.

The system is hosted on a GPU server at DSSC and is intended for use by the college’s faculty and students. 

According to the Headquarters Integrated Defence Staff, DRONA 2.0 has been trained on a large repository of academic material held by DSSC.

“DRONA 2.0 is pre-trained on a large repository of the academic content residing at DSSC providing major fillip to faculty, students of DSSC, aiding them towards access of information and progress advanced studies/research in the field of Defence and Strategic Studies,” the institute posted on X.

The platform supports reasoning, document analysis, image recognition and generation, speech-to-text, text-to-speech, translation, and transliteration. It can also generate podcasts and presentations and provide tools for data analytics.

DRONA was first introduced at DSSC in June 2025 as a customised GPT called DSSC Resource Optimising Neural AI, or DRONA. Version 1.0 was launched on the DSSC Wide Area Network on June 16, 2025.

The latest version moves beyond the earlier system by combining Sarvam 105B with open-source models and adding multimodal capabilities. 

DSSC said DRONA 2.0 was subjected to an audit and certified by the Headquarters Integrated Defence Staff and the Defence Cyber Agency before deployment.

Sarvam 105B is Sarvam AI’s flagship open-weight reasoning model. The company released the model, alongside Sarvam 30B, in March earlier this year at the India AI Impact Summit, with both models trained from scratch in India using compute provided under the IndiaAI Mission. 

The model supports 11 languages, including English and 10 Indian languages, and has a 128,000-token context window. Sarvam positions it for reasoning, coding, long-context document analysis, and agentic workflows.

The partnership with DSSC gives Sarvam another defence-sector application for its models. The company has said its AI systems are being adopted in sectors including banking, insurance, government technology, and defence.