Sarvam API pricing: DeepSeek at half its own peak-hour rate
Sarvam opened Token Factory to all developers on 16 November 2026, selling DeepSeek V4.1 Flash at ₹14 per million input tokens plus GST, at any hour. DeepSeek's own API charges about ₹28 in its peak hours, 6:30-9:30 am and 11:30 am-3:30 pm IST on weekdays.

Key points
- • DeepSeek's own API charges double from 6:30 to 9:30 am and from 11:30 am to 3:30 pm IST on weekdays.
- • Sarvam says DeepSeek requests sent through Token Factory are not passed on to DeepSeek's own API.
- • Every verified Sarvam account gets ₹100 of free credits, according to the [pricing docs](https://docs.sarvam.ai/api/getting-started/pricing).
Bengaluru's Sarvam AI has opened Token Factory, its service for running other companies' open AI models, to every developer, with Sarvam API pricing in rupees and servers in India.
The headline price is for DeepSeek V4.1 Flash, the Chinese model Sarvam calls "one of the most used models on OpenRouter", the API marketplace. Sarvam charges ₹14 per million input tokens and ₹55 per million output tokens on its Standard endpoint, plus GST. That rate holds at any hour.
That matters because DeepSeek's own API doubles its price during peak hours, and those hours fall in the middle of an Indian working day. Sarvam's introductory price lasts until 31 December 2026.
Sarvam announced the service in a blog post dated 16 November, saying Token Factory serves models "from GPU clusters located in India" through one OpenAI-compatible API. If you have used Sarvam's consumer app, our Sarvam AI app guide covers that side of the company.
DeepSeek's peak hours fall in the Indian working day
DeepSeek's own API charges double from 6:30 to 9:30 am and from 11:30 am to 3:30 pm IST on weekdays. Its pricing page sets peak hours of 01:00 to 04:00 and 06:00 to 10:00 UTC, Monday to Friday. Off-peak rates are half the peak rates, and weekends and Chinese public holidays are off-peak in full.
So a Bengaluru team running a support bot through the late morning pays DeepSeek's higher rate for most of its busiest hours. Sarvam's Standard price matches DeepSeek's off-peak rate almost exactly, and Sarvam lists no time-of-day pricing.
| DeepSeek V4.1 Flash, per 1M tokens | Input | Cached input | Output |
|---|---|---|---|
| Sarvam Standard (introductory, to 31 Dec 2026) | ₹14 | ₹0.25 | ₹55 |
| Sarvam Interactive (faster endpoint) | ₹25 | ₹0.50 | ₹105 |
| DeepSeek API, off-peak ($0.15 / $0.003 / $0.60) | about ₹14.09 | about ₹0.28 | about ₹56.36 |
| DeepSeek API, peak ($0.30 / $0.006 / $1.20) | about ₹28.18 | about ₹0.56 | about ₹112.71 |
Sources: Sarvam's API pricing docs and DeepSeek's Models & Pricing page, both read 17 Nov 2026. Sarvam's prices exclude GST. DeepSeek's dollar prices converted by BestAI at ₹93.93 to the dollar, from the European Central Bank's 16 Nov reference rates.
Add 18% GST, the usual rate on such services, and Sarvam's Standard price becomes about ₹16.52 for input and ₹64.90 for output. We could not confirm from DeepSeek's pages whether it adds Indian GST to its dollar bills.
Your prompts stay on Indian servers, Sarvam says
Sarvam says DeepSeek requests sent through Token Factory are not passed on to DeepSeek's own API. Its Token Factory page puts it plainly: "Requests are not forwarded to the original model developer's overseas API." Inference runs on Sarvam-operated GPUs in India, the same page says, and zero data retention is available on request.
That is Sarvam's own statement, not an audit. It covers only the model call: any outside tool your app connects to processes data wherever that tool runs.
Sarvam API pricing after the free ₹100 in credits
Every verified Sarvam account gets ₹100 of free credits, according to the pricing docs. At the Standard rate that buys about 7 million DeepSeek input tokens, or under 2 million output tokens.
After that you buy credits in rupees, one credit per rupee. Credits never expire, but Sarvam's FAQ says payments are "final and non-refundable", and there is a minimum purchase it does not state. Sarvam says it sends an invoice after every payment. The same key and credits work for Sarvam's speech, translation and document APIs.
Not every model is open to everyone. DeepSeek V4 Pro, Kimi K3.1 Pro and Qwen 3.8 27B run on provisioned throughput and are "enabled per organisation", so a solo developer can sign up and call DeepSeek V4.1 Flash, GLM 5.3, Gemma 4 31B and Sarvam 105B, but not those three.
Sarvam's speed claims are its own tests
Sarvam says its Standard endpoint streams DeepSeek V4.1 Flash at about 130 tokens per second and its Interactive endpoint at about 200, against 60 for DeepSeek's own API, in charts on its launch post. Sarvam says it ran its tests from an AWS server in Mumbai. Those are vendor-run tests, and we found no independent measurement.
What it means for you
If you run DeepSeek in production from India and most of your traffic lands between 6:30 am and 3:30 pm on weekdays, Sarvam's Standard endpoint halves the token price for those hours before GST, or cuts it by about 40% if you pay 18% GST and can't claim it back, until the end of December. Off-peak and weekend traffic costs about the same either way.
Switching is a two-line change: point an OpenAI-compatible client at https://api.sarvam.ai/v2, use your Sarvam key, and set the model to deepseekv4.1-flash-1017, Sarvam's docs show. Test output quality on your own prompts first; the model name is the same, but the serving stack is not.
Our DeepSeek review covers the model itself, and our Sarvam AI review covers the company's own models. If an outage on DeepSeek's API is what pushed you to look, our story on DeepSeek's five-hour API outage shows how it handled refunds.
Sarvam has not said what the Standard price will be from 1 January 2027.
Sources: Sarvam, "Introducing Sarvam Token Factory", 16 Nov 2026; Sarvam Token Factory models and FAQ; Sarvam docs, "Pricing, Credits & Rate Limits"; Sarvam docs, "Token Factory"; Sarvam API pricing and FAQ; DeepSeek API docs, "Models & Pricing"; European Central Bank euro reference rates, 16 Nov 2026.
How we reported this: we read Sarvam's launch post, its Token Factory, pricing and FAQ pages and DeepSeek's pricing page on 17 November 2026 between 5:30 and 6:30 pm IST, and converted DeepSeek's dollar prices ourselves. We have not run either API. The Sarvam pages we read do not say which payment methods it accepts, and DeepSeek's pricing page does not say whether it charges Indian GST.
Cover image: eight Nvidia GPU modules on an HGX B200 board on display, the GPU model Sarvam says it runs in India. Pokiiri, CC BY-SA 4.0, via Wikimedia Commons.
Source: https://www.sarvam.ai/blogs/sarvam-token-factory
Related tools
We may earn a commission if you buy through our links. This doesn't affect our rankings. Learn more about our affiliate disclosure.
Sarvam AI
Indian-language speech, translation and chat APIs billed in rupees, plus the free Indus app.
DeepSeek
Free Chinese AI chat and a very cheap API, with open weights you can host yourself.
See more of BestAI.in in Google's Top Stories: add us as a preferred source.