Sarvam ropes in Mistral founding team member Devendra Chaplot as adviser

Sovereign AI startup appoints former Mistral and xAI executive as adviser as it expands globally and builds a one trillion-parameter foundation model

Sovereign AI startup Sarvam has roped in Devendra Chaplot, a member of Mistral AI's founding team, as an adviser as it looks to scout for artificial intelligence (AI) talent globally. The announcement was made at Sarvam’s first developer conference, Epoch 2026, in Bengaluru.

 

Chaplot, who is based in Palo Alto, California, also worked at Thinking Machines Lab, the AI startup founded by former OpenAI Chief Technology Officer (CTO) Mira Murati. He worked there for a little more than a year before joining Elon Musk's xAI.

 

Chaplot’s appointment also comes at a time when Sarvam is setting up an office in San Francisco.

 

 

Addressing the audience at the conference, Chaplot said India needs to develop the capability to train its own models rather than depend on global frontier AI models, especially given the country’s linguistic, cultural and educational diversity, as well as its policy requirements. 

"At Sarvam, we want to build the best class models that are globally competitive, and part of this is about ensuring that the great minds who are building for various companies across the world, also build for India," Kumar said while introducing him.

 

Chaplot said a lot of people think building at the frontier requires a lot of resources, but that is not true. "there is a misconception that it takes a lot of experienced people which is also not true. You need a handful of them, mostly a lot of motivated and talented people and thats what we have plenty in India. We want to have models that are customised for Indian needs and that is why need to build models here," he added.

 

“The model is not the goal. The ability to build models is the goal,” Chaplot said, adding that developing expertise across research, infrastructure and data would allow companies, governments, developers and researchers in India to build and customise models for their own needs.

 

He also said building frontier-scale models in India is achievable with the right computing infrastructure. 

Sarvam also announced that it is building a one trillion-parameter foundation model in India. The company unveiled pricing for its flagship Sarvam 105B model. At $0.80 per one million blended tokens, it is among the most cost-effective offerings compared with $4.50 for OpenAI's GPT-5.4 Mini and $9 for Gemini 3.5 Flash.

 

The company said that for those building voice products, Sarvam offers the cheapest, fastest and most scalable solution.

 

Besides its language models, Sarvam also announced the commercial availability of Sarvam Vision Edge, its document intelligence platform.