Sarvam unveils plan for trillion-parameter AI model, expands voice and multimodal offerings

Indian AI startup Sarvam has outlined its most ambitious goal yet: building a trillion-plus parameter foundation model from scratch in India. The announcement was made at the company’s Epoch event on Thursday, where it also unveiled upgrades across its AI portfolio, including speech and vision models, AI agents, smart glasses and a service that lets developers rent Indian phone numbers.

The company said its next-generation model is being built to compete globally while serving enterprise and public sector use cases in India. “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 Pratyush Kumar, co-founder of Sarvam.

The planned model would be a significant step up from the company’s previously released open-weight models and forms part of its broader push to build AI infrastructure within the country.

Cutting the cost of enterprise AI

Sarvam also made a strong pitch around affordability, saying its latest models are substantially cheaper to run than competing frontier models. The company claimed inference on its models is 5.5 times cheaper than global rivals, arguing that reducing the cost of deploying AI applications is becoming just as important as building powerful models. “If you are building a voice product today, Sarvam is the cheapest, fastest and most scalable solution,” the company said.

It added that comparable voice offerings from ChatGPT are currently unavailable, while Gemini is not available at a similar scale. The company said lower inference costs would allow businesses to build and deploy AI applications more economically.

New models and enterprise tools

Alongside the roadmap for its frontier model, Sarvam announced upgrades to its multimodal AI stack.

Its Vision model has been enhanced to better understand documents, images and complex visual layouts. The company said the latest version improves visual reasoning and document intelligence while supporting Indian languages.

The speech platform also received major updates, including improvements to speech recognition, text-to-speech and conversational voice capabilities. According to the company, the models deliver lower latency, better accuracy and broader multilingual support, making them suitable for customer support, voice assistants and other enterprise applications.

Sarvam also showcased new AI agents and developer tools designed to help businesses build production-ready AI applications more easily.

The announcements reflect the company’s strategy of building domestically hosted AI infrastructure tailored for enterprises and public sector organisations.

Kaze smart glasses showcased

Sarvam also demonstrated its Kaze smart glasses, first introduced at the India AI Impact Summit earlier this year.

During the presentation, the company showed how the device could assist visually impaired users. Using onboard cameras, the glasses identified an approaching bus, provided information about its route, estimated the remaining travel time and alerted the user when their stop was nearing.

Phone number rental service launched

Among the more unusual announcements at the event was a service that allows users to rent Indian mobile numbers.

In a live demonstration, the company showed a user renting a phone number in about 30 seconds after completing PAN and Aadhaar verification.

The offering is aimed at developers building AI agents that need Indian phone numbers for voice calls and other communication-based applications.