Sarvam AI opens Token Factory inference service on Indian GPU clusters

Nov 17, 2026
Sarvam AI now serves its own 105B model and open-weight models such as DeepSeek, GLM, Kimi and Gemma from GPU clusters in India through one API.

Indian AI company Sarvam AI has opened Sarvam Token Factory to everyone. The service is the company's inference stack, serving its own models and leading open-weight models from GPU clusters located in India through one OpenAI-compatible API.

Alongside Sarvam's own 105B model, Token Factory serves GLM 5.3, GLM 5.3 Flash, DeepSeek V4.1 Flash, DeepSeek V4 Pro, Kimi K3.1 Pro and Gemma 4 31B. Sarvam says it runs India's largest cluster of Nvidia B200 GPUs and is expanding its fleet. It also runs the same inference stack on premises for what it calls critical institutions.

The company claims it matches or beats the leading inference providers on speed and cost for most of the models it serves. In its own tests, run from a client in AWS Mumbai over 72 hours, Gemma 4 31B returned its first token in 0.40 seconds on a 10,000 token prompt, against 2 seconds for the only other provider serving the model. Sarvam says it built the benchmark as a recreation of Artificial Analysis's method.

DeepSeek V4.1 Flash is offered through two endpoints at introductory prices: an interactive one at about 200 tokens per second per user, aimed at coding agents, priced at ₹25 per million input tokens and ₹105 per million output tokens, and a standard one for background work at ₹14 and ₹55. Sarvam says it serves more than 450 billion tokens a day on that model, with 95.7% of prompt tokens served from cache, according to its announcement.

We hope you enjoyed this article

Consider subscribing to one of our newsletters like Daily AI Brief.

Also, consider following us on social media:

Free newsletter

Daily AI Brief

Daily report covering major AI developments and industry news, with both top stories and complete market updates

Whitepaper

Tensordyne Napier: What If One Rack Could Do the Work of Nine?

Tensordyne

This Tensordyne whitepaper presents Napier, an inference-focused AI processor and rack-scale system based on the company’s TDN Math logarithmic number system. It examines infrastructure requirements for large mixture-of-experts and agentic models, compares major inference architecture approaches, and details the TDN AIP processor, TDN72 pod, TDN Link fabric, and Napier Ultra configuration. The paper reports simulation-based performance, cost, and accuracy-validation results, including Tensordyne’s projected comparison of one Napier rack with a nine-rack Nvidia Rubin plus Groq deployment; the chip is reported as taped out and in fabrication.

Read more
Free, six days a week

Daily AI Brief: the AI news that matters, in your inbox.