Sarvam AI launches Prism decision model in New Delhi

Sarvam AI has launched Prism in New Delhi, a decision-making model it says responds in 44 milliseconds on an NVIDIA B200 GPU and can be customised by enterprises using their own data.

Built on the company’s Sarvam 105B base model, Prism is aimed at structured, rapid tasks such as routing customer-support queries, analysing documents, conducting security checks and assessing AI-agent performance. Sarvam said the model averaged 71.8 across three benchmarks, its highest score among 25 decision models it tested that responded in under 100 milliseconds.

The company said Prism scored 82.5 on its internal Indian-language benchmark. In tests where the model gave answers with at least 90% confidence, Sarvam reported error rates of 2.2% for English responses and 4.4% for Indian-language responses.

Prism’s weights are available under the Apache 2.0 licence, which allows organisations to train and run the model on their own infrastructure. It is also available through Sarvam’s API.

Sarvam said its internal cybersecurity benchmark, covering 5,517 questions, gave Prism an accuracy score of 89%.