On October 22, Ant Digital Technologies officially released the decision-making model LingDT-Decision-1.0. Built specifically for rapid decision-making, the model offers higher accuracy, faster response and lower Token costs when handling structured decision-making tasks, making it especially suitable for industrial AI scenarios. LingDT-Decision-1.0 comes in two versions, 4B and 27B. The model is now available on Ant Digital Technologies' DTMaaS platform, and developers can try it for free for a limited time.
Testing by the Ant Digital Technologies team in real business scenarios such as customer service, risk control and finance shows that when handling the same tasks, LingDT-Decision-1.0 can reduce Token consumption by more than 90% and improve response time by 10 times compared with large language models, while delivering better performance.
Compared with other decision-making models, LingDT-Decision-1.0 pioneers native multi-select and sequential decision-making capabilities, making it better suited to real enterprise application scenarios. Across 23 open-source evaluation benchmarks, LingDT-Decision-1.0's overall score surpassed popular decision-making models such as Microsoft-Decision-1 and Jev 1.13., and the 27B version of LingDT-Decision-1.0 achieved the highest accuracy among models of the same parameter size.
(Caption: The advantage zones of the 4B/27B versions of LingDT Decision cover decision-making models of relevant sizes including Jev, MS Decision and GLiDE, demonstrating extremely strong overall cost-effectiveness)
Designed specifically for decision-making tasks, Token consumption cut by more than 90%, faster and more accurate
In actual enterprise applications, there are large numbers of high-frequency decision-making tasks.
For example, in a customer service scenario, after receiving an inquiry such as "This order has not shipped after three days, please handle it as soon as possible," the AI needs to quickly make a series of judgments and decisions: the type of issue (refund, shipping urging or complaint?), ticket urgency (whether it needs priority handling), the responsible department (which department the ticket should be routed to), and risk judgment (whether human takeover is needed).
Traditional large language models have obvious shortcomings in such scenarios. Large language models typically need to reason token by token (Token) and then generate output text, which means long chains, high costs, greater susceptibility to errors and unstable performance.
Decision-making models are designed precisely to address these pain points; their goal is not to generate text but to make fast, accurate judgments. They are therefore naturally suited to handling high-frequency decision-type tasks in enterprise business, such as intent detection, tool selection, rule classification and result scoring.
Currently, the LingDT-Decision-1.0 model is being tested in real business scenarios at Ant Digital Technologies, including intelligent customer service, security risk control and financial applications. Its core idea is to shift the large model from "first generating a structured answer, then parsing it afterward" to "making judgments directly among limited options." This not only reduces formatting problems caused by unstable output, but also makes the originally cumbersome parsing and validation process much simpler.
For example, in data annotation scenarios, compared with DS-V4-flash, Token consumption is reduced by about 97% and processing time is improved by about 10 times; in intelligent customer service scenarios, compared with Qwen-3.7-flash, Token consumption is reduced by 90% and processing time is improved by about 1 time, while risk identification accuracy rises from 76% to 89%.
In financial application scenarios, for tasks such as intent recognition, tool calling and hallucination detection, processing time is improved by about 5 times, and the full tool combination match rate is improved by about 2.6 times, while the hallucination recall rate rises from 44% to 70%, making risk detection more sensitive.
Overall, LingDT-Decision-1.0 consistently demonstrates the advantages of lower cost, faster response and more accurate judgment across multiple real-world scenarios. At the same time, it can use one set of models to handle multiple types of judgment tasks, making integration simpler and maintenance easier, so it is also easier to reuse and promote across different business scenarios.
Building the next generation of decision-making models with two world-first features
Compared with other decision-making models, Ant Digital Technologies' LingDT-Decision-1.0 places greater emphasis on being built for real enterprise application scenarios. In addition to basic capabilities such as judgment, single-select, classification and rating, LingDT-Decision-1.0 also pioneers native multi-select and sequential decision-making (DAG) features, better adapting to the complex, multi-step decision-making needs in enterprise business.
In enterprise scenarios, the more common situation is not "choose one" but "identify a set of items to handle and execute them in order." For example, when a user says: "I want a refund, cancel the renewal, and reissue an invoice, but don't close my account." A single-select model often only recognizes the main demand of "refund" and easily misses other needs. Multi-select can identify all demands at once; sequential decision-making can further judge that the refund requires verification first, canceling the renewal can be executed first, and invoice processing may need to wait for the refund result. Therefore, multi-select and sequential decision-making better match real business processes and can better improve accuracy and final handling outcomes.
For native multi-select capability, LingDT-Decision-1.0 allows a single task to hit multiple candidates simultaneously, fitting needs such as multi-label recognition, composite intents and tool combinations; for sequential decision-making capability, the model can complete multi-step decisions according to the dependencies between tasks, aligning more closely with real business processes in finance, risk control and customer service.
Test results covering 23 mainstream open-source evaluation benchmarks show that the 27B version of LingDT-Decision-1.0 ranks first in overall score (90% x single-select score + 10% x multi-select score), with accuracy surpassing popular decision-making models such as Microsoft-Decision-1, Jev 1.13.0 and Fastino/GLiDE. Among them, on multi-select tasks, the 4B and 27B versions of LingDT-Decision-1.0 achieve accuracy surpassing 16 decision-making models including Microsoft-Decision-1, Jev 1.13.0 and Star-lux, demonstrating the advantage of native multi-select capability in multi-label and combination judgment tasks. On single-select tasks, it ranks in the first tier.
(Caption: A 2-dimensional chart of single-select/multi-select capabilities shows that LingDT Decision 4B/27B belongs to Tier-1 in single-select & judgment capability, surpassing the vast majority of same-size models; while its "native multi-select capability" leads by a wide, discontinuous margin)
This article was provided by Ant and reprinted by QbitAI with authorization; the views belong to the original author.