Sarvam AI, a Bengaluru-based artificial intelligence startup, has launched two new large language models—Sarvam-30B and Sarvam-105B—marking a major step toward real-time AI applications and advanced reasoning capabilities.
With this launch, the company signals a strong push into enterprise-grade and sovereign AI infrastructure built within India. The announcement took place at the India AI Impact Summit 2026 in New Delhi, reinforcing the country’s ambition to develop homegrown foundational AI models aligned with national priorities.
Sarvam AI designed the Sarvam-30B model specifically for real-time conversational use. As a result, the model features a 32,000-token context window that supports longer, more coherent dialogues while maintaining low-latency inference. Consequently, the architecture makes the model well-suited for interactive applications, AI agents, and other latency-sensitive enterprise use cases where responsiveness remains critical.
At the same time, Sarvam AI positioned the Sarvam-105B model for heavier reasoning workloads and complex analytical tasks. The model offers a significantly larger 128,000-token context window, which enables multi-step problem solving, long-form analysis, advanced coding assistance, and research-driven workflows. Moreover, both models use a mixture-of-experts architecture that balances high performance with computational efficiency, allowing them to scale effectively across demanding environments.
Sarvam AI stated that it trained both models entirely from scratch, with a strong emphasis on Indian languages and efficiency. Therefore, the models aim to support enterprise deployments that require governance, analytics, multilingual interactions, and agent-based systems. Additionally, the company reported that the models demonstrate competitive performance against leading global counterparts across benchmarks for reasoning, coding accuracy, and language understanding.
This development closely aligns with India’s broader push for sovereign AI, where policymakers and industry leaders increasingly view locally developed models as critical national infrastructure. By training models on diverse linguistic and cultural data, Sarvam AI seeks to enable AI adoption across key sectors such as healthcare, education, governance, and industry. As a result, the launch positions the startup as a key contributor to India’s growing AI ecosystem while advancing the country’s goal of building globally competitive yet locally grounded AI systems.
