Sarvam AI: India's Trillion-Parameter Model Challenges OpenAI

Sarvam has announced plans to build a trillion-plus parameter foundation AI model in India. This move positions the company to compete with leading global AI labs. Sarvam co-founder Pratyush Kumar stated they are building this large language model from scratch in India. It aims to be competitive in areas like coding, cybersecurity, and scientific simulations. The company also updated its flagship Sarvam 105B model, which they claim outperforms OpenAI's GPT-5.4 Mini and Google's Gemini 3.5 Flash in certain benchmarks. What's interesting is Sarvam 105B's cost is significantly lower, at $0.80 per million blended tokens, compared to $4.50 for GPT-5.4 Mini. Beyond language models, Sarvam introduced Vision 2.0, an AI for document intelligence with improved support for handwritten Indian text. They also launched Sarvam Vision Edge, an on-device platform already used by the Odisha government for land records. The company plans to open a new office in San Francisco, signaling global expansion. This development could bring new competition and innovation to the global AI landscape.