Sarvam's new AI model makes decisions in just 44 milliseconds
What's the story
Sarvam AI, an Indian artificial intelligence (AI) company, has launched Prism, a cutting-edge decision-making model. The innovative system can respond in just 44 milliseconds and gives enterprises the flexibility to customize and deploy it on their own infrastructure. The launch is part of Sarvam's broader effort to integrate AI into business workflows more efficiently.
Model features
Prism achieved an average score of 71.8 across 3 benchmarks
Prism is built on Sarvam's 105B base model and is specifically designed for quick, structured decisions like customer support routing, document analysis, and AI agent evaluation.
In a blog post announcing the launch, Sarvam revealed that Prism achieved an average score of 71.8 across three benchmarks while responding in just 44 milliseconds on an NVIDIA B200 GPU.
This was the highest score among the 25 decision models tested by Sarvam that responded in under 100 milliseconds.
Model reliability
Model provided reliable confidence estimates
Prism is designed to provide reliable confidence estimates, a key requirement for enterprises looking to automate decisions.
In tests with answers given with at least 90% confidence, Prism made errors in just 2.2% of English responses and 4.4% of Indian language responses, lower than its own predicted error rates.
The model also gave consistent answers when the order of options was changed, further proving its reliability in decision-making processes.
Model accessibility
Prism's weights released under Apache 2.0 license
A major highlight of the launch is the release of Prism's weights under the Apache 2.0 license. This allows enterprises to train the model on their own data and run it on their own infrastructure.
The company said early applications of Prism include customer support routing, document intelligence, security checks, and evaluation of AI-agent performance.