Ant Digital Technologies, the enterprise technology arm of Ant Group, has released LingDT-Decision-1.0, a model built for structured business decisions rather than open-ended text generation. It comes in 4B and 27B parameter versions and is available on the company's DTMaaS platform, where developers can try it free for a limited time.
The idea is to split a common class of work out of general-purpose large language models. Many enterprise tasks are really a series of choices among known options. Ant Digital Technologies gives a customer service example: a message complaining that an order has not shipped after three days needs the system to decide whether it is a refund, a chase-up or a complaint, how urgent it is, which department should handle it, and whether a human needs to step in.
A general LLM usually reasons token by token, writes out text and then has that text parsed into a decision. The company says that chain is long and expensive, and it fails more easily. A decision model is designed to judge directly within a defined set of options. That makes it suited to high-frequency steps inside agent workflows, such as intent detection, tool selection, rule classification and scoring.
According to tests by Ant Digital Technologies in customer service, risk control and financial scenarios, LingDT-Decision-1.0 cut token consumption by more than 90% compared with large language models on the same tasks, responded up to 10 times faster and performed better. Beyond yes-or-no judgments, single choice, classification and rating, the company says the model is the first to support native multi-select and sequential decisions arranged as a directed acyclic graph (DAG), which fits multi-step business processes.
On 23 open-source benchmarks, the company says LingDT-Decision-1.0 outperformed other decision models, including Microsoft-Decision-1 and Jev 1.13, and that the 27B version achieved the highest accuracy among models of its size. These results are the company's own and have not been independently verified.
The release does not replace general LLMs. The model is meant to sit alongside them, handling the many small, repetitive judgments in an enterprise agent so that larger models can focus on open-ended reasoning and generation. For teams building agents at scale, that split could make a large difference to cost and latency.
SEE ALSO: Ant Digital Technologies Open-Sources Agentar-SQL · Ant Digital Technologies Launches AI Payment Feature on DTClaw