Sarvam AI Unveils Epoch Platform and Trillion-Parameter Model Plans

Bengaluru-based Sarvam AI launched the Epoch Builder Edition platform on July 30, 2026, designed to allow organizations to build and deploy large language models tailored for Indian languages and use cases. The platform supports over 10 Indian languages, including code-mixed variants like Hinglish and Kannanglish, and features 7-billion and 70-billion parameter models. Sarvam AI has already established partnerships with three Indian Institutes of Technology (IITs) and two state governments for pilots in education and public service delivery.

During its Epoch 2026 developer conference, the company announced it will develop a trillion-plus parameter foundational model from scratch within six months. This model aims to compete with frontier systems from OpenAI, Google, and Anthropic in specialized fields including cybersecurity, coding, simulation, and science. To support this goal, the startup hired former Mistral advisor Devendra Singh Chaplot and opened a new office in San Francisco to attract global talent.

Sarvam AI also introduced a suite of new products, including the Sarvam Code assistant, the Saaras V4 speech recognition model, and the Bulbul V4 text-to-speech model. The company is positioning its current 105-billion-parameter model as a cost-effective alternative to GPT-5.4 Mini, claiming a price of $0.80 per million blended tokens. These expansions follow a June funding round that secured $234 million of a $300 million Series B, valuing the company at $1.5 billion.