Indian AI company Sarvam AI has launched Sarvam Prism, a decision-making model aimed at enterprise workflows. The company says it can produce structured decisions with a median response time of 44 milliseconds when running on an NVIDIA B200 GPU, positioning it for use cases such as customer support routing, document analysis and evaluating AI agents.

Sarvam says Prism is built on its Sarvam 105B base model and reports performance results from its own benchmarks, including an average score of 71.8 across three tests. It also cites a higher score on an internal Indian-language benchmark and describes confidence estimates as an important feature for automation, stating that error rates in tests remain below predicted levels at multiple confidence thresholds.

Outlets frame the launch differently. Business Line leads with the enterprise automation promise and highlights speed, consistency and reliability. The Free Press Journal adds more detail from Sarvam’s claims, including benchmark scores, reported accuracy/error-rate findings, and the broader emphasis on deployment on customers’ infrastructure.