Live intelligence

Latest AI updates

1 updates
Products

Sarvam Launches Token Factory: India-Hosted Inference for 8 Open Models

Sarvam has opened Token Factory, a managed inference platform serving its own models and open-weight models from NVIDIA B200 clusters in India through one OpenAI-compatible API. The launch lineup lists 8 models including Sarvam 105B, GLM 5.3, DeepSeek V4.1 Flash 1017, Kimi K3.1 Pro and Gemma 4 31B, with on-premise deployment offered to enterprises. The stack combines a cache-aware control plane, separate prefill/decode GPU pools and a shared Mooncake-based cache; Sarvam reports 95.7% of prompt tokens served from cache for DeepSeek V4.1 Flash and a 2.2x throughput gain for Sarvam 105B from MXFP8 weights plus speculative decoding. Pricing starts at ₹10/₹25 per 1M tokens for Sarvam 105B (introductory through end of 2026), though all benchmarks are self-reported and Together still beats Sarvam on some DeepSeek streaming metrics.

Sarvam AILLM inference infrastructureIndia AI ecosystem