Sarvam Token Factory Opens India-Hosted AI Inference to All
Sarvam Token Factory serves Sarvam 105B, DeepSeek, Kimi, GLM and Gemma 4 from GPUs in India via one API, with rupee pricing and zero data retention.
Sarvam Token Factory serves Sarvam 105B, DeepSeek, Kimi, GLM and Gemma 4 from GPUs in India via one API, with rupee pricing and zero data retention.
NHA AI voice scribe: EkaCare will turn doctor-patient talk in 15+ Indian languages into FHIR notes for ABDM, running on air-gapped AWS infrastructure.
Sarvam AI launched Prism, a System-1 decision model that scores options in a median 44 ms, with a company-reported average of 71.8 across three indices.
DeepSeek’s Harness team has released Super-Agent, an MIT-licensed research prototype of a proactive coding agent that monitors GitHub and developer chats and works in parallel sessions, as it takes on Anthropic’s Claude Code.
Murai Labs has released Agaram 2.5B, an open Tamil–English language model trained from random weights with a tokenizer built for Tamil, on a pretraining compute bill of about USD 4,765. It translates better than Sarvam-1 in the directions tested but remains weak on general knowledge.
The Central Institute of Indian Languages, Mysuru, has released Anuvani-LID and Anuvani-Diarization, open-weight speech AI models for language identification and speaker diarisation across all 22 scheduled languages and Indian English.