Sarvam AI Enters Decision Model Market with Prism

Sarvam AI has launched Prism, a model built on its 105B-parameter foundation model, optimized for specific decision-making tasks. Rather than generating text, Prism processes inputs to return probabilities for a defined set of options, aiming to handle tasks such as customer support routing, fraud detection, and document classification more efficiently than general-purpose large language models.

Technical benchmarks provided by the company indicate a median response time of 44 milliseconds on Nvidia B200 hardware. Sarvam has released the model under an Apache 2.0 license, allowing enterprises to host it on their own infrastructure. The company asserts that this approach balances performance with the need for data privacy and control within enterprise environments.

This move highlights a shift in the Indian AI ecosystem, where startups are increasingly focusing on specialized models to compete with global general-purpose offerings. Competitors such as ConvAI Innovations and Sthānika AI are also targeting this space with distinct technical priorities. Industry observers note that while benchmark scores are relevant, the long-term viability of these models will depend on their reliability and ability to integrate into complex enterprise workflows.