India Unveils Sarvam AI Model for Language Processing

In a significant development for India's artificial intelligence (AI) landscape, Sarvam AI has launched two advanced language models during the AI Impact Summit in New Delhi. These models, which aim to improve capabilities in Indian languages, were unveiled by Pratyush Kumar, the co-founder of Sarvam AI, on February 18, 2026.

The announcement comes in the wake of the Indian government's initiative to create its own large language models, similar to those developed globally, including ChatGPT. Both models released by Sarvam AI are open source, which aligns with the government's emphasis on making technology accessible.

One of the models showcased at the summit featured 30 billion parameters, and during its demonstration, it referred to itself as Vikram. This name pays homage to the Indian space pioneer Vikram Sarabhai. However, Sarvam AI later clarified that a definitive name for the model has yet to be confirmed.

Kumar highlighted the competitive edge of Sarvam AI's language models, claiming they outperformed other leading models currently in use. Developing such models is a resource-intensive process, requiring substantial investment in graphics processing units (GPUs) that work collaboratively in large data centres. Nonetheless, the Ministry of Electronics and Information Technology in India has directed its focus more on model inference—utilising pre-trained models for queries—rather than on the training of new models.

Following the emergence of the Chinese language model DeepSeek R1, Indian IT Minister Ashwini Vaishnaw stated that Indian companies would also be capable of training their own language models. Sarvam AI has secured significant funding from private investors, including Peak XV and Khosla Ventures, who have collectively contributed over $50 million towards its development. In addition, Sarvam AI gained subsidised access to computing power through the IndiaAI Mission’s common compute programme.

A primary goal of the Sarvam AI models is to enhance performance in Indian languages. Existing AI models have struggled with Indian languages due to a lack of available text resources for training purposes. Kumar showcased the models' translation capabilities, demonstrating their proficiency in translating English into various Indian languages.

While the models have generated considerable anticipation, they are not yet publicly available. Sarvam AI’s website mentions that a chat feature will be introduced soon, signalling their commitment to enhancing user interaction with the AI technologies they have developed.