Sarvam launches Token Factory inference platform in India
Start here
Sarvam has launched Sarvam Token Factory, an India-hosted inference platform that serves eight models through a single OpenAI-compatible API, including its own Sarvam 105B and open-weight models such as GLM 5.3, DeepSeek V4.1 Flash, Kimi K3.1 Pro and Gemma 4 31B, running on NVIDIA B200 clusters in India. The stack is built on vLLM, SGLang, Dynamo and Mooncake; Sarvam says caching serves 95.7% of prompt tokens on DeepSeek V4.1 Flash, which it says handles more than 450B tokens per day. Pay-as-you-go models run on shared endpoints, while DeepSeek V4 Pro, Kimi K3.1 Pro, Qwen 3.8 27B, Clef and EmbeddingGemma 2 require Provisioned Throughput enabled per organisation; on-premise deployment is offered to enterprises. Introductory pricing for Sarvam 105B starts at ₹10 per 1M input tokens and ₹25 per 1M output tokens through the end of 2026. Sarvam's benchmark figures are self-reported, and Together AI remains faster on DeepSeek V4.1 Flash streaming.
Generated from reports · updated 14 h ago
Report timeline
Follow the reports to see the event from different sides.
- MarkTechPostSarvam launches Token Factory, an India-hosted inference platform serving 8 open models
Sarvam launched Sarvam Token Factory, an inference platform that serves its own Sarvam 105B and open-weight models such as GLM 5.3, DeepSeek V4.1 Flash 1017, Kimi K3.1 Pro and Gemma 4 31B from NVIDIA B200 clusters in India through one OpenAI-compatible API. The stack builds on vLLM, SGLang, Dynamo and Mooncake, and Sarvam reports that caching serves 95.7% of prompt tokens on DeepSeek V4.1 Flash, which handles 450B+ tokens per day. Introductory pricing starts at ₹10 input and ₹25 output per 1M tokens for Sarvam 105B through the end of 2026, and all benchmark numbers are self-reported, with Together AI still faster on DeepSeek V4.1 Flash streaming.
Heat for this event
- Current, comparable range
- 7
- Peak, comparable range
- 10November 17 18:00
- Change in 24 hours
- –
The trend compares the same participants over hours observed continuously, so its range can be narrower than the current heat count. Hover or tap the chart for hourly heat; the arrow keys move through it.