Indian AI startup Sarvam AI has opened its Token Factory platform to developers and businesses, offering access to its own AI models as well as popular global models such as DeepSeek, Kimi and Google's Gemma through a single API. The platform, which works somewhat like OpenRouter, lets users run different AI models without setting up separate infrastructure for each, with the processing handled on GPU clusters located in India.Sarvam says Token Factory is designed to make AI model access faster and more affordable for Indian users, while giving businesses greater control over where their data is processed.
The paid service charges users in rupees based on the number of tokens they consume.Indian alternative to OpenRouter and Hugging Face Token Factory works somewhat like global AI platforms such as OpenRouter and Hugging Face, which allow developers to access and use different AI models through a common platform. While OpenRouter lets users run multiple AI models through a single API, Hugging Face offers a large library of AI models that developers can explore and use.
Sarvam is positioning Token Factory as an India-based alternative, going a step further by providing the computing infrastructure to run these models on GPU clusters located in the country.For developers and businesses, this means they can choose an AI model based on their needs without having to set up or manage the infrastructure required to run it. They can also switch between models as their requirements change, without being tied to a single AI provider.
Sarvam says it has made this possible through a single OpenAI-compatible API. Developers who already use OpenAI's API can connect their applications to Token Factory by changing the API endpoint and selecting the model they want to use.
They can access all supported models using the same Sarvam API key, software development kit and credits.The platform's launch catalogue includes Sarvam's own Sarvam 105B model, alongside global open-weight models such as GLM 5.3, GLM 5.3 Flash, DeepSeek V4.1 Flash, DeepSeek V4 Pro, Kimi K3, Kimi K3.1 Pro and Google's Gemma 4 31B. Sarvam's wider catalogue also lists Qwen 3.8 27B, an embedding model and its Prism decision model.
What sets Token Factory apart is that Sarvam hosts the models on its own GPU infrastructure in India, rather than forwarding prompts to overseas servers operated by model providers. This, the company says, can help reduce response times while giving Indian businesses greater control over where their data is processed.
Sarvam also revealed that it operates India's largest cluster of Nvidia B200 GPUs and is expanding its computing capacity to support more AI models and growing demand.How much does Sarvam Token Factory cost?Interestingly, unlike many AI services that quote prices in US dollars, Sarvam has listed its Token Factory prices in Indian rupees. The company will charge users based on the number of tokens processed by a model.
Tokens are essentially small pieces of text that an AI model reads or generates. The more a user interacts with a model, the more tokens it consumes.Sarvam will charge Rs 10 for every million input tokens and Rs 25 for every million output tokens used with its Sarvam 105B model.
This is the introductory pricing. For DeepSeek V4.1 Flash, the company is offering two options.
The Standard option, meant for background tasks, will cost Rs 14 per million input tokens and Rs 55 per million output tokens. The Interactive option, designed for coding and tasks that need faster responses, will cost Rs 25 and Rs 105, respectively.Prices for other models will vary.
For instance, Sarvam will charge Rs 200 per million input tokens and Rs 950 per million output tokens for Kimi K3.1 Pro. All prices exclude GST.- Ends

