Sarvam AI is the clearest example of an Indian AI company that ships rather than announces. It was founded in Bengaluru in August 2023 by Vivek Raghavan and Pratyush Kumar, who previously worked on AI4Bharat at IIT Madras, and raised roughly 41 million dollars in a combined seed and Series A in December 2023 led by Lightspeed with Peak XV Partners and Khosla Ventures. In April 2025 MeitY selected it under the IndiaAI Mission to build an indigenous foundation model, with access to government-supported GPU capacity.
The model line runs from Sarvam-1 in October 2024 through Sarvam-M in May 2025 under Apache 2.0 to Sarvam-30B and Sarvam-105B, both announced on 18 February 2026. The 105B is a mixture-of-experts model with a 128,000 token context window; the 30B is a mixture-of-experts model with about 1 billion active parameters per token and a 32,000 token window. Both were open-sourced under Apache 2.0 in early March 2026 with weights on Hugging Face and AIKosh. A consumer app built on the 105B, called Indus, reached beta on iOS, Android and web on 20 February 2026.
Beyond the language models it ships Saaras V3 for speech across Indian languages, Sarvam Vision for document understanding and OCR in Indian scripts, and announced Sarvam Kaze, AI wearable glasses supporting more than ten Indian languages, for May 2026. A Sarvam Startup Program launched in March 2026 offering selected early-stage companies six to twelve months of API credits and engineering support.
Pricing is published openly, in rupees and excluding GST. Sarvam 105B costs 15 rupees per million input tokens, 5 for cached input and 60 for output. Speech to text is 0.49 rupees a minute, text to speech with Bulbul v4 Flash is 1 rupee per thousand characters, translation is 2 rupees per thousand characters, document digitisation is 0.50 rupees a page and dubbing is 40 rupees per minute of source media. Pay as you go carries no monthly fee and verified users get 100 rupees in free credits. Sarvam also resells third-party models through the same API, including Gemma 4 31B at 35 and 90 rupees, DeepSeek V4.1 Flash at 25 and 105, GLM-5.3 at 126 and 390, and Kimi K3 at 270 and 1,350 rupees per million input and output tokens.