Bengaluru-based Sarvam AI has launched Sarvam Prism, a new AI model built for fast business decisions — answering in just 44 milliseconds — with open weights that enterprises can run on their own infrastructure.

What Prism does differently

  • Launched on 3 November 2026, Prism is a “System One” decision model, named after the fast, intuitive mode of human thinking
  • Instead of generating text word by word like a chatbot, it reads a situation, a question and a set of predefined options in a single forward pass and picks the best one
  • It is built on Sarvam 105B, the company’s 105-billion-parameter base model, using LoRA fine-tuning
  • It can handle long inputs of up to 131,072 tokens in one pass — useful for judging long documents or AI agent activity logs

Speed and scores

  • Median response time of 44 milliseconds on a single NVIDIA B200 GPU
  • Average score of 71.8 across three benchmark indices — 67.0 on the Jev Decision Index, 65.9 on Sarvam’s recreation of the Artificial Analysis Decision Index, and 82.5 on an Indian-language benchmark covering 10 Indian languages and Hinglish
  • Sarvam says it was the highest-scoring model under 100 milliseconds among the 25 models it tested
  • These are Sarvam’s own reported results, not independently verified

Open weights for enterprises

  • Prism’s weights are released under the Apache 2.0 licence on Hugging Face, so companies can train it on their own data and run it on their own servers
  • It is also available through Sarvam’s API on Sarvam Indus
  • Running it needs one GPU with at least 120 GB of memory
  • Typical uses include customer-support routing, document analysis, content moderation and evaluating AI agents

Related articles on HimachalGovt