tech
February 18, 2026
Indian AI lab Sarvam's new models are a major bet on the viability of open-source AI
The new lineup includes 30-billion and 105-billion parameter models; a text-to-speech model; a speech-to-text model; and a vision model to parse documents.
TL;DR
- Sarvam AI launched new open-source large language models (LLMs) to compete with larger global rivals.
- The models include 30B and 105B parameter versions, text-to-speech, speech-to-text, and vision capabilities.
- These models utilize a mixture-of-experts architecture for reduced computing costs and support large context windows.
- They are trained from scratch on trillions of tokens, including multiple Indian languages, and designed for local use cases.
- Sarvam aims to scale thoughtfully, focusing on relevant real-world applications and plans to open-source the 30B and 105B models.
- The initiative aligns with India's goal to reduce reliance on foreign AI platforms.
- Funding comes from investors like Lightspeed Venture Partners and Khosla Ventures, with support from the IndiaAI Mission, Yotta, and Nvidia.
Continue reading
the original article