Sarvam AI model: 30B, 105B models unveiled in India, aims to bring AI assistant to feature phones

On Day 3, Sarvam AI has unveiled two latest AI models during the AI Impact Summit 2026: Sarvam 30B and Sarvam 105B. The highlight of this event was both the models support Indian languages, such as Punjabi, Marathi, Hindi and more.

Co-founder Pratyush Kumar, demonstrates how both the models are trained to serve the population at a reasonable price. Let’s delve deeper into what are these models and how will it float in India.

Sarvam 30B model: Conversational AI for feature phones

The Sarvam 30B model targets a broad set of use cases spanning multilingual text generation, conversational AI, translation and content creation. At 30 billion parameters, it is designed to be efficient yet powerful, suitable for deployment in scenarios where both performance and accessibility matter.

During the launch, Kumar showcased several benchmarks showing that Sarvam 30B has already defeated a few famous AI models.

After all the jargons, Sarvam AI finally came to the exciting part of use case. During the event, the company demonstrated how this model can help in real-life situation. It showed that with this model, even a feature phone (now a rare breed with keypad) can be used for hold a conversation with AI.

More from Tech

Elon Musk’s xAI engineer sparks debate after boasting about 19-hour work shift Coders, translators...10 jobs at highest risk as AI reshapes labour market, says Anthropic

Sarvam AI aims to serve the local people in the country and Sarvam 30B fits exact with this description.

In the demo, using as an example, with just a phone call, the feature phone was turned into a high-end device that can reply in several Indian languages, including Punjabi and Hindi.

Sarvam 105B: For more complex tasks

The Sarvam 105B model, the more powerful of the two, is engineered for complex reasoning and high-level language tasks. With a context window of 128K tokens, it can process and retain far larger amounts of information than most comparable systems, enabling it to handle sophisticated use cases such as long-form summarisation, in-depth analytical queries, and nuanced dialogue generation.

According to Sarvam AI, the model’s scale and efficiency allow it to outperform global peers, including the DeepSeek R1 model, marking a significant leap for India’s AI capabilities on the international stage.

During the AI Impact Summit 2026, Sarvam demonstrated the model’s real-world potential through an impressive use-case example.

The team uploaded a company’s balance sheet, and within seconds, the 105B model analysed the data and began answering complex financial and contextual questions with precision.

Quick Reads

View All

Best phones under Rs 60,000 (March 2026): Vivo X200T, OnePlus 13s to Motorola Signature

Elon Musk’s xAI engineer sparks debate after boasting about 19-hour work shift

The demonstration showcased the system’s ability to process structured and unstructured data simultaneously, proving how Indian-built AI can now deliver enterprise-grade analytical intelligence across sectors such as finance, governance, and business operations.