AI & Models
Sarvam bets on open-source AI to reduce reliance on foreign platforms
Sarvam has unveiled new open-source AI models, aiming to reduce reliance on foreign platforms while betting that efficient, smaller systems can compete with U.S. and Chinese rivals.
On Tuesday, Indian AI lab Sarvam unveiled a new generation of large language models. The launch, which took place in New Delhi, aligns with New Delhi’s push to reduce reliance on foreign AI platforms. The startup is betting that smaller, efficient open source AI models will be able to grab market share from larger U.S. and Chinese rivals. The new lineup includes 30-billion- and 105-billion-parameter models, marking an upgrade from the company’s 2-billion-parameter Sarvam 1 model released in October 2024.
The new models utilize a mixture-of-experts architecture—an architecture that activates only a fraction of total parameters at a time to reduce computing costs. Sarvam plans to open source the 30-billion- and 105-billion-parameter models. The specifications for these models include:
- The 30-billion-parameter model supports a 32,000-token context window and was pre-trained on about 16 trillion tokens of text.
- The 105-billion-parameter model offers a 128,000-token context window and was trained on trillions of tokens spanning multiple Indian languages.
The models were trained using computing resources provided under the IndiaAI Mission, which is a government-backed AI initiative. To support the training and deployment of these models, the startup received infrastructure support from data center operator Yotta and technical support from Nvidia.
Founded in 2023, Sarvam has raised more than $40 million in total funding. Its investors include Lightspeed Venture Partners, Khosla Ventures, and Peak XV Partners. Rather than focusing on raw size, the company plans to take a measured approach to scaling. “We don’t want to do the scaling mindlessly. We want to understand the tasks which really matter at scale and go and build for them,” said Sarvam co-founder Pratyush Kumar.
Why it matters
Sarvam’s strategy highlights a growing trend of sovereign AI development, where local labs build specialized, efficient models to bypass reliance on dominant U.S. and Chinese tech stacks.