Sarvam-105B Is Here: India’s Biggest AI Leap Yet?

India’s push to build homegrown AI LLMs took a major step forward on 18 February, as AI startup Sarvam AI launched two new large language models. These models are called Sarvam-30B and Sarvam-105B The company said both models will be released as open source, making them accessible to developers, enterprises, and government agencies.

The launch comes at a time when India is trying to reduce its dependence on foreign AI systems and strengthen its own digital infrastructure, especially in areas like data security, cost control, and language support.

What Can Sarvam-30B and Sarvam-105B Be Used For?

Sarvam AI introduced Sarvam-30B as a smaller and more efficient model designed to deliver strong output while using fewer tokens. The company said the model was pre-trained on 16 trillion tokens and supports a context length of 32,000 tokens. This makes it suitable for tasks such as customer support, document summaries, and enterprise workflows.

The bigger model, Sarvam-105B, was positioned as the flagship system. It supports a much larger context window of 128,000 tokens, allowing it to handle long documents, detailed conversations, and complex business use cases. Sarvam AI said this model performs on par with other large systems of similar scale, including both open-source and proprietary models.

What is Special About Sarvam-30B and Sarvam-105B?

A key part of Sarvam AI’s message was that the models were developed from the ground up without relying on external datasets. This is especially relevant in India’s current AI environment, where data ownership and sovereignty are becoming increasingly important.

With the government-backed IndiaAI Mission supporting the development of sovereign AI infrastructure, the launch fits into a broader national effort to build foundation models locally.

Real-World Demonstration Highlights India Focus

Along with benchmarks, Sarvam AI also showcased real-world use cases. One of the most notable demonstrations involved a chatbot running on a basic feature phone with a physical keypad. The chatbot, named “Vikram”, was shown conversing in multiple Indian languages, including Hindi and Punjabi.

This highlighted Sarvam AI’s focus on building systems that work not just for premium users, but also for mass-market India.

Read more: Best Artificial Intelligence Stocks for February 2026: Persistent Systems, Saksoft and More Based on 5 Year CAGR.

Conclusion

With Sarvam-105B and Sarvam-30B, Sarvam AI has made a bold statement in India’s growing AI race. By releasing both models as open source and focusing on long-context capabilities and Indian language use cases, the startup has positioned itself as a serious player in India’s mission to build sovereign, scalable AI systems. 

Disclaimer: This blog has been written exclusively for educational purposes. The securities mentioned are only examples and not recommendations. This does not constitute a personal recommendation/investment advice. It does not aim to influence any individual or entity to make investment decisions. Recipients should conduct their own research and assessments to form an independent opinion about investment decisions.