Understand this story
On November 22, 2026, Sarvam AI released the open-weight model checkpoint Sarvam-105B-1711 for customer service, voice agents, and agentic tasks, offering both BF16 and 65 GB NVFP4 weights. Sarvam AI said the NVFP4 version achieved an average score of 73.9 across eight conversational-agent benchmarks, with blended pricing of $0.15 per million tokens, and can run on a single RTX PRO 6000 Blackwell GPU. The model’s context length has been extended to 160K tokens; the weights are now available on Hugging Face, and the API is accessible through Sarvam Token Factory. There are currently no further developments supported by sufficient factual evidence.
AI-generated from public reports
Latest development · Report
Sarvam AI Releases Sarvam-105B-1711 Open-Weight ModelDevelopments
1 confirmed developments
Sarvam AI Releases Open-Weight Model Checkpoint Sarvam-105B-1711
Source:Sarvam AI Blog
Reports and discussion timeline
Browse reports and discussion by time. Their evidence levels remain distinct.
Nov 22
Sarvam AI BlogFirst-partySelected
Sarvam AI Releases Sarvam-105B-1711 Open-Weight Model
Sarvam AI has released the open-weight Sarvam-105B-1711 model for customer service, voice agents, and agentic tasks, offering BF16 and 65 GB NVFP4 weights. Its NVFP4 version achieved an average score of 73.9 across eight conversational agent benchmarks, with blended pricing of $0.15 per million tokens, and can run on a single RTX PRO 6000 Blackwell GPU. The model’s context window has been extended to 160K tokens, the associated weights are now available on Hugging Face, and the API is available through Sarvam Token Factory.
AI Trends