AI unicorn Sarvam AI opens Token Factory API to run global AI models in India
Developers and enterprises can now use Sarvam AI's Token Factory, which gives them one API for the company's own models and a set of global ones.

Developers and enterprises can now use Sarvam AI’s Token Factory, which gives them one API for the company’s own models and a set of global ones. Sarvam said the models run on GPU clusters located in India.
Sarvam 105B sits on the platform next to open-weight models including Google’s Gemma 4 31B, DeepSeek V4.1 Flash, GLM 5.3 and Kimi K3.1 Pro. Through an OpenAI-compatible API, developers can move between the supported models without building their own computing infrastructure.
Pricing follows usage, counted in the tokens a customer sends to a model and the tokens the model produces in return. Sarvam’s catalogue puts Sarvam 105B at ₹15 per million input tokens and ₹60 per million output tokens, while cached input tokens cost ₹5 per million. Token Factory also sells dedicated GPU capacity, and Sarvam deploys its inference software on premises for critical institutions.
Sarvam claimed that for most of the models it hosts, the platform matches or beats leading global inference providers on speed and cost. In the company’s own tests, Sarvam 105B delivered its first token in 200 milliseconds and generated 330 tokens per second for a single user. Sarvam also said it serves more than 450 billion tokens a day on DeepSeek V4.1 Flash, and that 95.7% of prompt tokens come from cache.
With the launch, Sarvam’s business now covers running models for customers as well as building them. Others already operate in that market: Krutrim began hosting DeepSeek models on Indian servers in January 2025, and OpenAI offers India-based API inference for eligible GPT-6 Luna workloads. Anthropic provides in-country inference for select Claude models through Amazon Bedrock.
Sarvam will also compete with managed inference providers such as Together AI and Fireworks AI, as well as cloud providers including Amazon Web Services, Microsoft Azure and Google Cloud.
Sarvam’s pitch to enterprises is that they can use several models through one API and keep prompt processing in India. Requests stay on its own infrastructure and are not sent on to the overseas APIs of the model developers. For banks and other financial institutions that must process certain workloads in India, local hosting could help address data residency requirements.





