Hyderabad AI Lab Sthānika Says Its Sieve-31B-v3 Model Ties For First On Global Decision Benchmark

Sthānika, the Hyderabad-based AI research lab of the PurpleTalk group, says its open-weight decision model Sieve-31B-v3 has tied for first place on the Decision Index, an open benchmark hosted on Hugging Face for AI models built to make quick decisions. The model scored 71.1 in the board’s 28 October 2026 update.

Instead of generating text, Sieve models return a probability for each answer option, which the lab says makes decisions faster, cheaper and easier to audit than general-purpose language models. The model runs on a single 96 GB GPU. PurpleTalk was co-founded by Sridhar Muppidi.

#Sthanika #IndianAI #Hyderabad #Startup77