Sarvam launches token factory hosting 8 open AI models on India GPUs

What happened
Sarvam launched the Sarvam Token Factory on November 16, 2026, providing an inference platform that hosts eight open-weight AI models on NVIDIA B200 GPU clusters located in India. The platform offers access to these models through an API compatible with OpenAI's format and is available to both developers and enterprises. It processes over 450 billion tokens daily and supports multiple deployment options, including serverless and provisioned throughput. The service includes features for caching, optimization, and data privacy, with no data retention and local inference capabilities.
Key details
- Pricing for Sarvam 105B starts at ₹10 per million input tokens and ₹25 per million output tokens, with introductory rates through the end of 2026.
- The platform serves over 450 billion tokens each day.
- Inference runs on Sarvam-operated GPUs in India, not routed to overseas APIs.
- The service guarantees 99.95% uptime for enterprise deployments and does not retain customer data.
Why it matters
Developers and enterprises in India now have access to high-throughput, low-latency AI model inference hosted locally, which can reduce reliance on overseas infrastructure and address data residency requirements. Organizations benefit from flexible deployment models, competitive pricing, and contractual uptime guarantees, while maintaining data privacy due to zero data retention and local processing. This enables Indian companies and institutions to integrate advanced AI capabilities with more control over costs, performance, and compliance.

