Sarvam AI Releases Prism Decision Model with 44 ms Latency on 120 GB GPU

Sarvam AI introduced Prism, a LoRA fine‑tune of its 105 billion‑parameter model that delivers responses in 44 milliseconds and scores an average 71.8 on three benchmarks, outperforming 24 other decision models with sub‑100 ms latency. The model runs in FP8 precision on a single GPU with at least 120 GB memory and is available via Hugging Face and the Sarvam API.

Benchmarks are the company's own; not independently verified.