Sarvam Plans Trillion-Parameter AI Model, Launches India-Hosted Inference Service

AI startup Sarvam has announced its most significant product roadmap to date, revealing plans to build a trillion-plus parameter frontier AI model within India. The company also launched a domestically hosted inference platform, aiming to create an end-to-end AI stack encompassing models, infrastructure, and enterprise software.

The Bengaluru-based startup unveiled these initiatives at its inaugural developer conference, Epoch 2026. The event also featured several new launches across agentic AI, speech, vision, and enterprise productivity tools.

"We are very happy to announce that we are building a trillion-plus parameter model right here in India. We are building them from scratch to be competitive in coding, cybersecurity, simulation, science and more," said Sarvam co-founder Pratyush Kumar during the conference. A specific timeline for the model's release was not disclosed.

Sarvam also announced the opening of an office in San Francisco and the appointment of Devendra Singh Chaplot, who was part of the founding teams at Mistral AI and Thinking Machines Lab, as an advisor.

A key launch was Sarvam Inference, a platform that serves frontier open-source AI models using infrastructure hosted within India. The service currently supports Sarvam's own 105 billion parameter model, alongside open models like GLM 5.2 and Gemma 4. Sarvam stated the platform aims to help developers and enterprises access AI models while ensuring inference workloads remain within India, addressing growing data residency requirements.