Skip to main content

Sarvam launches Token Factory for AI model inference

17 NOVEMBER 2026·2 MIN READ·5 SOURCES

Sarvam AI launched Token Factory on November 16, 2026, describing it as an inference platform for its own and open-weight AI models. The service is aimed at developers and enterprises and is open to everyone, Sarvam says.

Sarvam launches Token Factory for AI model inference

Key takeaways · 3

  • 01

    Developers already using OpenAI’s API can connect applications by changing the endpoint and choosing a model, according to India Today.

  • 02

    Sarvam’s introductory Sarvam 105B price is ₹10 per million input tokens and ₹25 per million output tokens; listed prices exclude GST.

  • 03

    Sarvam says zero data retention is available on request, and its SLA guarantees 99.95% uptime.

A shared inference platform

Sarvam describes Token Factory as an inference platform that serves its own models and open-weight models from GPU clusters in India.[2] The launch catalogue includes Sarvam 105B, GLM 5.3, GLM 5.3 Flash, DeepSeek V4.1 Flash 1017, DeepSeek V4 Pro, Kimi K3.1 Pro and Gemma 4 31B.[2] The company says the service is open to everyone, while Unite.ai reports that it is aimed at developers and enterprises.[2][1] Sarvam positions the offering as an India-based alternative providing computing infrastructure to run models on GPU clusters in India.[3]

One API, model switching

Sarvam says users can access the listed models through one OpenAI-compatible API.[2] India Today reports that developers already using OpenAI’s API can connect their applications by changing the API endpoint and choosing a model.[3] The publication also reports that Token Factory lets users switch between models without setting up or managing the infrastructure needed to run them.[3] For teams assessing a migration, the stated endpoint change and model selection are the practical integration details; the evidence does not specify further migration requirements.

Usage-based prices

Sarvam charges according to token usage, with prices listed in Indian rupees.[3] Its introductory price for Sarvam 105B is ₹10 per million input tokens and ₹25 per million output tokens.[3] Sarvam separately lists introductory DeepSeek V4.1 Flash pricing at ₹14 per million input tokens and ₹55 per million output tokens.[2] The company says listed prices exclude GST.[4] These figures provide starting points for cost comparisons, but the evidence does not establish how a particular workload would perform or what it would cost overall.

Location, controls and claims

Sarvam says requests run on its own GPUs in India rather than being forwarded to model developers’ overseas APIs, and says the platform is designed to give businesses greater control over where data is processed.[5][3] Zero data retention is available on request under an arrangement in which prompts and responses are not retained, according to Sarvam.[5] Sarvam also says the service guarantees 99.95% uptime under an SLA.[4] Its claims that it matches or beats leading global providers on speed and cost for most models, and its published performance comparisons, are company statements rather than independent findings in the evidence.[2][1]

For teams evaluating model access, the combination of a shared API, usage-based pricing and India-based inference may make it easier to compare deployment options. Buyers should validate costs, data-handling terms and performance against their own workloads before choosing a service.

Why it matters
For teams

AI training for your team

kju.ai is built for teams: daily sessions, team analytics, and rollout support for your organization.

How this developed

  1. 17 November 2026

    Sarvam launches Token Factory for AI model inference

Sources

AI fluency, one session a day, built for your work.