IT之家 · 智能时代 · 10/22/2026, 18:12:04
Ant Digital Launches LingDT-Decision-1.0, Cutting Token Costs by Over 90%
Ant Digital released LingDT-Decision-1.0, a model optimized for high-frequency decision tasks with native multi-select and sequential decision capabilities. It reduces token consumption by over 90% and improves response speed by 10x compared to LLMs in enterprise scenarios, achieving top accuracy among same-parameter models.
SOURCE COVERAGEFull Original Report
According to the introduction, compared to other decision models, LingDT-Decision-1.0 is the first to support native multi-selection and sequential decision-making capabilities, making it better suited for real-world enterprise application scenarios. Across 23 open-source benchmark evaluations, LingDT-Decision-1.0 outperformed popular decision models such as Microsoft-Decision-1 and Jev 1.13. The 27B version of LingDT-Decision-1.0 achieved the highest accuracy among models with comparable parameter counts.
Tests conducted by the Ant Digital Technologies team in real business scenarios—including customer service, risk control, and finance—show that when handling identical tasks, LingDT-Decision-1.0 can reduce token consumption by over 90% compared to large language models (LLMs). It also improves response latency by 10x while delivering superior performance.
In actual enterprise applications, there are numerous high-frequency decision-making tasks. For example, in a customer service scenario, upon receiving an inquiry like, "This order hasn't been shipped after three days; please handle it urgently," the AI needs to quickly perform a series of judgments and decisions: determining the issue type (refund, urging shipment, or complaint?), assessing ticket urgency (does it require priority handling?), assigning the responsible department (which team should receive the ticket?), and evaluating risk (is human intervention necessary?).
Traditional LLMs exhibit significant shortcomings in these scenarios. LLMs typically reason token-by-token before generating output text, resulting in long processing chains, high costs, increased susceptibility to errors, and unstable performance. Decision models are specifically designed to address these pain points. Their goal is not to generate text but to make fast and accurate judgments. Consequently, they are naturally well-suited for high-frequency decision-oriented tasks in enterprise workflows, such as intent recognition, tool selection, rule-based classification, and result scoring.
Compared to other decision models, Ant Digital Technologies' LingDT-Decision-1.0 places greater emphasis on being built for real-world enterprise application scenarios. Beyond foundational capabilities such as judgment, single-choice selection, classification, and rating, LingDT-Decision-1.0 introduces native multi-selection and sequential decision-making (DAG) features for the first time. These capabilities better accommodate complex, multi-step decision requirements common in enterprise operations.
LingDT-Decision-1.0 is available in two versions: 4B and 27B. The model is currently live on Ant Digital Technologies' DTMaaS platform, where developers can experience it for free for a limited time.