Sarvam.ai releases Sarvam-105B-1711 Flash-Lite model
Sarvam.ai has released Sarvam-105B-1711, an open-weight Flash-Lite model trained on B200 infrastructure and optimized for customer support, voice agents, and agentic tasks. The company reports that the model improves instructions following, tool use, and Indian-language performance while cutting its blended price to $0.15 per million tokens. Additionally, the weights are made available on Hugging Face in BF16 and an NVFP4 version capable of running locally on a single workstation GPU. Open article →