Synopsis: Bengaluru-based Sarvam AI has opened Token Factory, an India-hosted service that runs its own 105B model and global open-weight models through one API. EQMint maps the launch against Sarvam’s fast 2026 timeline, from a $1.5 billion valuation in June to a 30-fold revenue jump, and its own speed claims versus rivals.
Sarvam AI, the Bengaluru startup founded in 2023 by Pratyush Kumar and Vivek Raghavan, has opened its Token Factory inference platform to all developers and enterprises, in a blog post dated 16 November 2026.
Token Factory lets customers call Sarvam’s own Sarvam 105B model and a set of open-weight models from other developers, including Google’s Gemma 4 31B, DeepSeek V4.1 Flash, DeepSeek V4 Pro, GLM 5.3 and Kimi K3.1 Pro, through a single OpenAI-compatible API. All of them are served from GPU clusters in India. Sarvam says it runs “India’s largest B200 cluster”, a claim EQMint could not independently check.
What Sarvam has launched
The pitch is data residency. Requests stay on Sarvam’s own GPUs in India instead of going to overseas model APIs, which matters for banks, insurers and government bodies. Sarvam also offers dedicated GPU capacity and on-premise deployment.
Introductory prices for DeepSeek V4.1 Flash are ₹14 per million input tokens and ₹55 per million output tokens on the standard endpoint, and ₹25 and ₹105 on the faster interactive endpoint. Sarvam says this model alone serves more than 450 billion tokens a day, with 95.7% of prompt tokens served from its three-tier cache. For its own Sarvam 105B model, the company reports a first token in 0.2 seconds and 330 tokens per second for a single user.
EQMint analysis: Sarvam’s 2026 timeline and its own speed claims
Token Factory comes five months after Sarvam became a unicorn. On 15 June 2026 it announced a $234 million first close of a $300 million Series B at a $1.5 billion post-money valuation, led by HCLTech with $150 million. HCLTech’s exchange filing put its stake at 10.46% for ₹1,427.3 crore. On 24 July, HCLTech, Sarvam and the Odisha government announced an AI data centre in Bhubaneswar with a planned outlay of ₹14,257 crore. Sarvam’s FY26 revenue was ₹45.1 crore (unaudited) against ₹1.5 crore in FY25, a roughly 30-fold rise by EQMint’s calculation.
Sarvam also published output speeds for DeepSeek V4.1 Flash across providers. These are its own measurements, not independent tests.
| Provider / endpoint | Output speed (tokens per second) |
|---|---|
| Together | 278 |
| Sarvam Token Factory, interactive | 200 |
| Fireworks | 193 |
| Baseten | 165 |
| Sarvam Token Factory, standard | 130 |
| GPT-6 Luna (reference) | 67 |
| DeepSeek API | 60 |
Source: Sarvam AI blog “Introducing Sarvam Token Factory” (company-reported benchmarks), figures as of 16 November 2026.
By Sarvam’s own numbers, its interactive endpoint is about 3.3 times the DeepSeek API’s speed but trails Together. Its pitch rests on in-country hosting as much as on raw speed.
Who it competes with
Sarvam enters a crowded field. Krutrim has hosted DeepSeek models on Indian servers since January 2025, OpenAI began India-based API processing for eligible GPT-6 Luna workloads on 12 November, and Anthropic offers in-country inference for some Claude models through Amazon Bedrock, Inc42 reported. Critics have asked why a self-styled sovereign AI company hosts foreign open-weight models; Sarvam says local hosting is what delivers data residency.
The launch follows other recent Indian AI deals, including Bharat Semi’s $24 million seed round and Dreamteam’s $53 million raise.
What it means
For founders building AI products in regulated sectors, Token Factory is another way to keep data in India without owning GPUs. For Sarvam, it turns heavy compute spending into usage-based revenue. The company has not disclosed customer numbers for the service; pricing on other hosted models and progress on the Odisha data centre are the next markers.
Image credit: AICreate / EQMint (representational).
Sources
- Source: Sarvam AI, Introducing Sarvam Token Factory
- Source: HCLTech, Sarvam raises $234 million in first close of $300 million Series B
- Source: HCLTech, AI Data Center in Bhubaneswar with Sarvam and Government of Odisha
- Source: Inc42, Sarvam AI opens Token Factory to run global AI models on Indian servers
- Source: Inc42, Sarvam enters unicorn club with $234 Mn round led by HCLTech
- Source: Inc42, Sarvam’s AI inference play
Disclaimer: This article is for information purposes only and is not investment advice.
For more such information, visit EQMint
Join our WhatsApp channel for timely updates: Whatsapp


