Bengaluru-based artificial intelligence startup Sarvam AI is making a bold bet on a massive 1-trillion-parameter foundation model, signaling its ambition to compete with global AI leaders. The announcement came during the company’s Epoch 2026 event, where co-founder Pratyush Kumar outlined the model’s focus on coding, cybersecurity, scientific research, and simulation workloads.
Developed from scratch, the model aims to enable developers and enterprises to build AI solutions tailored specifically for Indian languages and use cases. At the same event, Sarvam AI revealed pricing for its earlier Sarvam 105B model: $0.80 per one million blended tokens—significantly cheaper than comparable global large language models (LLMs).
To bolster its efforts, the startup announced partnerships with three Indian Institutes of Technology (IITs) and two state governments for pilot projects in education and public service delivery. Additionally, the company has brought on board Devendra Singh Chaplot, a former xAI researcher, as a part-time advisor. Chaplot previously contributed to founding teams at Mistral AI and Mira Murati’s Thinking Machines Lab.
However, industry observers question whether building ever-larger models remains the right strategy. The AI landscape has shifted toward efficiency, with many companies adopting architectures like Mixture-of-Experts (MoE) to deliver strong performance with lower computational costs. A large model alone will not guarantee success—Sarvam must also develop practical tools that solve real-world problems for users and businesses.
While hiring a former xAI researcher and launching the 1-trillion-parameter project are significant steps, the startup’s biggest challenge will be turning these ambitious ideas into trusted, everyday products.
Leave a Reply