The Indian Army has developed a multimodal artificial intelligence system powered by Sarvam AI’s 105-billion-parameter model, marking another step toward building sovereign AI capabilities for defence applications. The system has been developed around the Defence Services Staff College (DSSC) academic repository and is designed to work with different types of information, including documents and images, while also supporting speech, translation and research functions.
The development is significant because India’s armed forces are increasingly experimenting with AI systems that can operate on sensitive military and institutional data. Rather than relying exclusively on general-purpose foreign AI platforms, the Army’s use of an Indian foundational model points toward a broader push for domestic AI infrastructure, multilingual capabilities and greater control over defence data.
Indian Army Uses Sarvam 105B For Multimodal AI
At the centre of the system is Sarvam-105B, a large language model developed by Indian AI startup Sarvam AI.
The Army’s system is reportedly trained on the DSSC’s academic repository, giving the model access to defence-related educational and research material relevant to the institution. The resulting system is not limited to text-based question answering.
It supports multiple AI capabilities, including document analysis, image generation, speech tools, translation and research.
Indian Army AI System At A Glance
| Parameter | Details |
|---|---|
| Organisation | Indian Army |
| AI model | Sarvam-105B |
| Model developer | Sarvam AI |
| Model size | 105 billion parameters |
| Training material | DSSC academic repository |
| AI type | Multimodal |
| Document analysis | Yes |
| Image generation | Yes |
| Speech capabilities | Yes |
| Translation | Yes |
| Research assistance | Yes |
| Key objective | Defence-focused AI capability |
The system’s multimodal design means users can work with different forms of information rather than being restricted to conventional text prompts.
What Is Sarvam-105B?
Sarvam-105B is one of Sarvam AI’s large foundational models developed specifically with India’s linguistic and technological requirements in mind.
The model has 105 billion parameters, putting it among the largest Indian-developed open AI models.
Sarvam has also developed a smaller Sarvam-30B model. Both models were designed to support Indian-language AI applications, with the broader Sarvam model family supporting 22 Indian languages, according to information published about the company’s defence-focused AI initiative.
Sarvam AI Model Stack
| Model | Parameters | Positioning |
|---|---|---|
| Sarvam-30B | 30 billion | Smaller foundational model |
| Sarvam-105B | 105 billion | Larger foundational model |
Sarvam AI Model Sizes
Comparison of the parameter counts of Sarvam-30B and Sarvam-105B.0306090120Sarvam-30BSarvam-105B
Parameter count refers to model size, not a direct measure of model quality or performance.
Sarvam-105B therefore has 3.5 times as many parameters as Sarvam-30B, although parameter count alone does not determine how capable an AI system is.
The Army’s System Is Multimodal
A key feature of the Army’s AI system is its ability to handle multiple types of information.
Traditional large language models primarily work with text. Multimodal systems can process or generate different forms of content, potentially including text, images and speech.
The Indian Army system reportedly combines document analysis with image generation, speech capabilities, translation and research tools.
Multimodal AI Workflow
Document
↓
AI Analysis
↓
Context Extraction
↓
Research / Reasoning
↓
Text, Speech Or Image Output
This can make the system more useful in institutional and defence environments where information is not stored exclusively as plain text.
Why The DSSC Repository Matters
The AI system has reportedly been trained on the academic repository of the Defence Services Staff College.
That is important because generic AI models are trained on broad datasets and may not have the specific institutional context required for specialised military education or research.
Training or adapting an AI system around an organisation’s own knowledge base can make it more useful for domain-specific tasks.
Generic AI Vs Defence-Specific AI
| Generic AI | Army AI System |
|---|---|
| Broad knowledge | DSSC-focused knowledge |
| General users | Defence users |
| General-purpose answers | Institutional research |
| Public information | Organisation-specific repository |
| Broad language capabilities | Defence-focused context |
The approach is similar to how businesses are increasingly deploying private AI systems trained or grounded in their own internal knowledge bases.
Potential Uses For Defence Education And Research
The system’s connection to the DSSC repository suggests that one major application could be supporting military education and research.
AI could potentially help personnel locate relevant material, summarise lengthy documents, compare research topics, translate material or assist with academic work.
The exact operational uses of the system have not been publicly detailed, so its capabilities should not be interpreted as evidence that it is being used for battlefield decision-making or classified intelligence operations.
Potential Applications
| Application | Potential Role |
|---|---|
| Document analysis | Search and summarise large documents |
| Research | Assist with academic research |
| Translation | Convert material between languages |
| Speech | Voice-based interaction |
| Image generation | Generate visual material |
| Knowledge retrieval | Access institutional information |
| Education | Support military learning |
The currently reported deployment is therefore best understood as a specialised defence AI system, rather than an autonomous battlefield AI.
Why Sovereign AI Matters For The Indian Army
Defence organisations have particular concerns around data security, sovereignty and external dependencies.
Sensitive military information cannot always be sent to public cloud-based AI services operated by foreign companies.
A domestically developed model can potentially give Indian institutions greater control over how AI systems are deployed and where their data is processed.
Sovereign AI Model
Indian AI Model
↓
Indian Infrastructure
↓
Army Knowledge Base
↓
Controlled Deployment
↓
Defence AI Applications
This approach can reduce dependence on external AI platforms for certain specialised workloads.
However, the precise security architecture of the Army’s system has not been publicly disclosed.
India’s Defence AI Push Is Expanding
The Army’s Sarvam-based system is part of a wider push toward artificial intelligence across India’s armed forces.
Information published earlier this year indicated that more than 75 AI projects were active across India’s armed services. Other initiatives have included AI systems for surveillance, research and military applications.
India’s Defence AI Landscape
| Area | AI Application |
|---|---|
| Army | Knowledge and multimodal AI |
| Navy | Maritime surveillance and monitoring |
| Air Force | Intelligence and operational support |
| Education | AI-assisted research |
| Logistics | Potential optimisation |
| Surveillance | Image and sensor analysis |
| Language | Translation and information access |
The growing number of projects suggests that AI is gradually moving from experimentation toward more specialised defence applications.
Sarvam AI Is Also Building Defence-Focused Infrastructure
Sarvam AI has separately positioned itself for government and defence applications through Chanakya, an applied-AI initiative designed for environments requiring on-premise and air-gapped deployment.
According to information published about the initiative, Chanakya is designed to combine Sarvam’s models with multimodal processing and agentic workflows for environments where public-cloud deployment may not be appropriate.
Sarvam’s Defence AI Strategy
Sarvam-30B
Sarvam-105B
↓
Chanakya
↓
On-Premise / Air-Gapped AI
↓
Defence + Government Applications
The Army’s use of Sarvam-105B therefore fits into a wider effort by the Indian AI company to develop domestic AI infrastructure for sensitive environments.
IndiaAI Mission Provides The Compute Foundation
India’s broader sovereign-AI push has also been supported by the government’s IndiaAI Mission.
Sarvam AI was among organisations selected to develop indigenous foundational models under the programme. Information published about the company’s defence strategy says Sarvam received government support through the IndiaAI Mission and access to subsidised compute.
The IndiaAI Mission has an overall outlay of ₹10,372 crore and has been used to expand access to AI computing infrastructure in India.
India’s AI Infrastructure Push
| Metric | Reported Figure |
|---|---|
| IndiaAI Mission outlay | ₹10,372 crore |
| GPUs onboarded | 38,000+ |
| Sarvam foundational models | 30B + 105B |
| Indian languages supported by Sarvam stack | 22 |
This infrastructure is important because training and operating large AI models requires significant computing capacity.
Why 105 Billion Parameters Matter
A parameter is a numerical value learned during AI model training that helps determine how the model processes information.
A 105-billion-parameter model is substantially larger than many conventional AI systems.
However, bigger does not automatically mean better.
Performance also depends on training data, model architecture, optimisation, inference techniques and the specific task being performed.
What Model Size Can Influence
More Parameters
→ Potentially greater capacity
↓
More Complex Representations
↓
Potentially Better Performance On Some Tasks
But:
Model Size ≠ Guaranteed Accuracy
The Army’s choice of Sarvam-105B therefore reflects the model’s overall capabilities and suitability for the intended deployment, rather than simply its parameter count.
Multilingual AI Could Be Particularly Valuable
India has a highly multilingual environment, making language technology an important part of sovereign AI.
Sarvam’s models have been developed with support for 22 Indian languages, according to information about the company’s model and defence strategy.
For a large organisation such as the Indian Army, multilingual capabilities could help personnel interact with AI systems across different linguistic environments.
Multilingual AI Potential
English
↕
Hindi
↕
Regional Indian Languages
↓
Translation
↓
Knowledge Access
This could be particularly useful for training, document translation and information retrieval.
AI Could Reduce Time Spent Searching Military Documents
One of the most practical applications of a system trained on an institutional repository is information retrieval.
Instead of manually searching through thousands of documents, an AI system can potentially locate relevant information and summarise it.
Traditional Research
Search Documents
↓
Read Multiple Sources
↓
Identify Relevant Information
↓
Create Summary
AI-Assisted Research
Ask Question
↓
Search Repository
↓
Retrieve Relevant Material
↓
Generate Summary
↓
Human Reviews Output
The final human-review step remains important because AI systems can generate incorrect or incomplete information.
Human Oversight Remains Critical
The Army’s adoption of AI does not mean that AI-generated information should automatically be treated as authoritative.
Large language models can produce inaccurate information, misunderstand context or generate plausible but incorrect answers.
This becomes particularly important in defence environments.
AI + Human Model
AI
Searches
Analyses
Summarises
Translates
↓
Human
Verifies
Interprets
Decides
↓
Final Output
The reported system’s academic and research orientation makes this human-in-the-loop approach particularly relevant.
The Bigger Picture
The Indian Army’s use of Sarvam-105B represents an important step in India’s effort to develop sovereign, domain-specific AI capabilities for defence. The system reportedly combines a 105-billion-parameter Indian foundation model with the DSSC’s academic repository and multimodal functions covering documents, images, speech, translation and research.
The development also illustrates how India’s AI strategy is moving beyond simply creating large language models. The next stage is deploying those models inside specialised organisations, using proprietary knowledge and controlled infrastructure. For the Army, that could eventually mean AI systems designed around military education, research, information retrieval and other specialised workflows while maintaining greater control over sensitive data.
Looking Ahead
The immediate significance of the Army’s Sarvam-105B system lies in its move toward a more specialised and locally controlled AI capability. As India’s armed forces develop more AI projects, the focus is likely to expand from document and research assistance toward areas such as intelligence analysis, logistics, surveillance and decision-support systems. The exact applications will depend on security requirements and operational validation.
For Sarvam AI, the Army deployment could also demonstrate the commercial and strategic value of Indian foundation models in high-security environments. If the technology proves reliable at scale, similar deployments could emerge across other government departments and defence institutions. The larger goal is not simply to build an Indian alternative to foreign AI models, but to create AI systems that can operate within India’s own data, language, security and infrastructure requirements
Get the day’s top stories in your inbox
One concise email. No spam, unsubscribe anytime.
