Archive · 20 Nov 2025
India's First Indigenous Large Language Model by Sarvam.ai in Early 2023
Sarvam.ai, an Indian startup, is going to create the country's first large language model.
This model will understand Indian languages better than others because it's trained on more Indian data.
This is a big step for India to have its own AI technology.
Sarvam.ai to launch India's first foundational large language model in early 2023
Model to have 120 billion parameters, trained on 17 trillion tokens with 15-20% Indian-origin data
Significant increase from current open-source models where Indian data comprises less than 1%
India AI Mission selected Sarvam as the first startup to build India’s foundational AI model
Launch represents a major milestone in India’s efforts to develop sovereign AI capabilities
- Who
- Sarvam.ai, an Indian startup led by Vivek Raghavan
- What
- Launching India's first foundational large language model
- When
- Early 2023
- Where
- India
- Why
- To develop sovereign AI capabilities and improve understanding of Indian languages
Pro-Indigenous Development
Open-Source Advocates
Indian Data Concentration
Pro-Indigenous Development
Important for better understanding of Indian languages and culture
Open-Source Advocates
Potential bias and lack of global exposure
- Model Size
- 120 billion parameters
- Training Data
- 17 trillion tokens, 15-20% of Indian-origin
- Expected Launch
- Early 2023
- Selection
- First startup chosen by India AI Mission for this task