Indian AI Firm Sarvam Open-Sources Large Reasoning Models
Indian AI firm Sarvam has open-sourced two large-scale reasoning models with 30 billion and 105 billion parameters. The release is expected to democratize access to advanced AI and spur innovation by allowing developers and researchers to build on the state-of-the-art models freely.
The release of these models is a significant step in India's push for "sovereign AI," aiming to reduce reliance on foreign-developed artificial intelligence systems. Sarvam AI is one of the key startups supported by the Indian government's ₹10,372-crore IndiaAI Mission, which provided the necessary computing power for training these large-scale models. The newly released models utilize a "Mixture-of-Experts" (MoE) architecture, which enhances computational efficiency. The 30-billion-parameter model is designed for real-time conversational applications, while the more powerful 105-billion-parameter version is intended for complex, multi-step reasoning tasks. Both were trained from the ground up on data that includes 22 Indian languages. On performance benchmarks, the 105B model has shown strong results in agentic reasoning and task completion, outperforming other models like DeepSeek R1 and Gemini 2.5 Flash in some tests. The models' weights are accessible on platforms like Hugging Face and AIKosh under an Apache 2.0 open-source license, allowing for broad use and development. Internally, Sarvam is already deploying these models. The 30B model powers their conversational AI platform, "Samvaad," while the 105B model is the foundation for the "Indus" AI assistant, which is designed for more complex workflows. This move is intended to provide Indian enterprises and developers with powerful, locally relevant AI tools. Sarvam AI was co-founded in August 2023 by Dr. Vivek Raghavan, who has a background in building India's digital public infrastructure, and Dr. Pratyush Kumar, a former researcher at Microsoft and IBM. In a significant funding round for an Indian AI startup, the company raised $41 million in its Series A, led by Lightspeed Venture Partners with participation from Peak XV and Khosla Ventures. The company has also entered into a collaboration with Microsoft. This partnership aims to make Sarvam's AI models, particularly those focused on Indian languages, available to a wider audience through Microsoft's Azure cloud platform, further integrating Indian-developed AI into the global ecosystem.
