Sarvam AI Launches Prism Decision Model for Enterprise Workflows

Indian artificial intelligence company Sarvam AI has introduced a new decision-making model named Sarvam Prism, aimed at expanding the use of AI in business workflows. Built on the company's Sarvam 105B base model, Prism is designed to handle fast and structured decisions, including customer support routing, document analysis, and the evaluation of AI agents.

According to Sarvam, the model delivered responses in 44 milliseconds on an NVIDIA B200 GPU and recorded an average score of 71.8 across three benchmarks. The company stated that this was the highest score among 25 decision models it tested that responded in under 100 milliseconds.

In internal benchmark testing, Prism recorded a score of 82.5 on Sarvam's Indian-language benchmark. By comparison, tests showed Perplexity Decider v1.2 scored 75 and Jev 1.14 scored 70.8. On an internal cybersecurity benchmark covering 5,517 questions, Prism achieved an accuracy rate of 89 per cent, compared to 84.5 per cent for Clef and 78.6 per cent for Jev 1.14.

The company noted that the model is designed to provide reliable confidence estimates for automated decision-making. In tests involving answers given with at least 90 per cent confidence, Prism reported error rates of 2.2 per cent for English responses and 4.4 per cent for responses in Indian languages, which the company stated were below its predicted error rates. Additionally, Sarvam reported that Prism maintained consistent answers across 2,000 questions when options were arranged in a fixed order.

A notable aspect of the launch is the release of Prism's weights under the Apache 2.0 licence. This licensing structure allows enterprises to train the model using their own data and deploy it on their proprietary infrastructure. Alternatively, the model is also accessible through Sarvam's API.

Early identified use cases for the Prism model include customer support routing, document intelligence, security checks, and the performance evaluation of AI agents.

"The launch of Sarvam Prism highlights the growing focus on domain-specific efficiency and infrastructure control within the enterprise AI segment. By offering weights under an Apache 2.0 licence and enabling local deployment, Sarvam AI is directly addressing data privacy and customisation requirements that businesses prioritize. Low latency response times and reliable confidence estimates are critical operational factors for companies looking to integrate AI deeper into structured workflows like document analysis and customer support routing." — Dr. Shishir Gupta, Founder & CEO, StartupLanes