Ant group's LingDT-Decision-1.0 model reduces token usage by over 90%

On October 22, 2026, Ant Group's Ant Numerical Science released LingDT-Decision-1.0, a decision-making model designed for rapid industrial applications. The model excels in structured decision tasks with higher accuracy, faster response times, and reduced token costs. It outperformed popular models like Microsoft-Decision-1 and Jev 1.13 in 23 open-source benchmarks, with its 27B version achieving the highest accuracy among models of similar size. In real-world testing across customer service, risk control, and finance, LingDT-Decision-1.0 reduced token consumption by over 90% compared to large language models, while improving response speed by tenfold. The model supports native multi-choice and sequential decision-making (DAG), making it suitable for complex, multi-step business decisions. Ant Numerical Science offers two versions, 4B and 27B, on its DTMaaS platform, with a limited-time free trial available for developers.