Sakhanda Wire
NVDA $153.77 -1.52% MSFT $504.91 -2.53% GOOGL $281.52 -0.78% META $620.99 -0.87% AMZN $256.80 -0.37%
← Back to the news

Sarvam opens India-hosted inference platform to all developers

Sarvam has made Sarvam Token Factory available to everyone. The managed inference platform serves Sarvam and open-weight models from NVIDIA B200 GPU clusters in India, using one OpenAI-compatible API. The eight models listed at launch include Sarvam 105B, GLM 5.3, GLM 5.3 Flash, DeepSeek V4.1 Flash 1017, DeepSeek V4 Pro, Kimi K3, Kimi K3.1 Pro and Gemma 4 31B. Enterprises can also request on-premise deployment.

The system combines a cache-aware router, separate GPU pools for prompt processing and decoding, and a cache using GPU memory and a shared G3 store built on Mooncake. Sarvam says caching supplied 95.7% of prompt tokens for DeepSeek V4.1 Flash, which processes more than 450B tokens daily. MXFP8 weights and speculative decoding reduced the Sarvam 105B requirement from two B200 GPUs per replica to one, according to the company.

Performance figures are self-reported. Sarvam measured 0.51 seconds to the first token from Mumbai for DeepSeek V4.1 Flash with a 1K prompt, but Together reported 0.71 seconds with a 10K prompt versus Sarvam’s 0.90 seconds. Listed introductory prices through the end of 2026 start at ₹10 input and ₹25 output per 1M tokens for Sarvam 105B.

Written by Sakhanda Wire AI from reporting by MarkTechPost. Facts, quotes and full context are in the original article below.

Originally published by MarkTechPost on

Read the original on MarkTechPost ↗

The reporting and the image belong to MarkTechPost and are credited and linked above; the text on this page is Sakhanda Wire's own summary.

← Back to the news

More stories

All the latest news