Ant Digital Technologies Debuts Decision Model With Benchmark Lead

Ant Digital Technologies introduced LingDT-Decision-1.0 for business decision tasks, with the 27B edition taking first place by combined score in evaluations spanning 23 open-source benchmarks.

Its accuracy exceeded that of Microsoft-Decision-1, Jev 1.13.0 and Fastino/GLiDE, and led among models of comparable size. The model comes in 4B and 27B versions on Ant Digital Technologies’ DTMaaS platform, where developers can try it free for a limited time.

Instead of generating text for a separate system to parse, LingDT-Decision-1.0 selects from preset options. It can also make multiple selections and carry out decisions in sequence based on task dependencies.

Tests across customer service, risk management and finance showed token use more than 90% lower and response times improved by up to 10 times. In customer service, risk-identification accuracy increased from 76% to 89%. In finance, hallucination recall rose from 44% to 70%.

TokenPost has also covered Liquid AI’s d1-ultra decision model, which assigns probabilities to supplied answer choices.