Sarvam AI · Decision Model

Prism

Prism is a LoRA fine-tune of Sarvam 105B that returns a probability score for each option in a single forward pass, rather than generating tokens. Sarvam reports an average of 71.8 across three benchmarks, including 65.9 on its recreation of the Artificial Analysis Decision Index, and a 44 ms response time on an NVIDIA B200 GPU. The weights are on Hugging Face under Apache 2.0 and can run in FP8 on a single GPU with at least 120 GB of memory, and the model is also available through the Sarvam API.