Sarvam Co-founder Pratyush Kumar
India’s full-stack sovereign AI company Sarvam AI on Thursday unveiled plans to build a trillion-plus parameter foundational model and launched a suite of new AI products and developer tools as it sharpens its ambitions to become a full-stack AI platform.Speaking at the company’s Sarvam EPOCH event, Co-founder Pratyush Kumar said the company is scaling its model-building capabilities to compete with leading global AI companies.“We are investing in trillion-plus parameter models from scratch to be competitive across coding, cybersecurity, simulation, science and other tasks,” he highlighted, adding that the model is likely to be ready in the next six months. The company said it plans to significantly expand its compute infrastructure and training capabilities, building on its experience of training large models in India.As part of its global expansion strategy, Sarvam had earlier announced the opening of a San Francisco office to tap into the world’s largest AI talent pool. “We want to have a very strong conduit of talent back to India. It is important to get the best people thinking and building for the Indian context,” Kumar commented.Pricing strategyAccording to Sarvam, its 105-billion-parameter model delivers competitive performance with leading open and closed-source frontier models of a similar class across knowledge, reasoning and agentic benchmarks. The company said the model also outperforms peers on Indian-language benchmarks while demonstrating strong capabilities in mathematics, coding and instruction following.The company also highlighted the pricing of its latest models, saying its 105-billion-parameter model is priced at $0.80 per one million blended tokens, compared with $4.50 for GPT-5.4 Mini and $9 for Gemini 3.5 Flash, offering a cost advantage of more than 5x and 11x, respectively.The start-up unveiled a range of products spanning enterprise AI, developer tools and multimodal AI capabilities. Among the launches were Sarvam Code, an AI coding assistant; Saaras V4, a multilingual speech recognition model; Bulbul V4, a multilingual text-to-speech model; and broader availability of its Voice Agents platform for developers.Kumar added that the company’s strategy extends beyond building frontier models to creating an end-to-end AI ecosystem. “We want to produce the largest share of AI tokens consumed in India right here.” He added that the goal is to enable developers and enterprises access frontier AI models hosted in India with strong data security and globally competitive pricing.ValuationThe announcements follow Sarvam’s $234 million first close of its $300 million Series B round in June, which valued the company at $1.5 billion post-money. Sarvam builds across the AI stack: training and inference infrastructure, frontier model research and a go-to-market motion spanning enterprises, developers and government. The investment will fund Sarvam’s continued research on training its next frontier model for agentic, coding and cybersecurity use cases, as well as access to compute at scale to expand its forward-deployed motion across key verticals.Published on July 30, 2026
