LingDT-Decision-1.0
Ant Digital Technologies released LingDT-Decision-1.0 in 4B and 27B sizes, a decision model that picks among defined options directly and uses over 90% fewer tokens than LLMs, by its own tests.
It makes structured choices directly within a set of options instead of writing out text that is then parsed into a decision. The company says it is the first to support native multi-select and sequential decisions arranged as a directed acyclic graph. It is meant to handle high-frequency steps in agent workflows such as intent detection, tool selection, rule classification and scoring, next to general LLMs.
- Date
- Friday 23 October 2026
- Lab
- Ant Digital Technologies
- Kind
- model
- Access
- closed API
- Price
- Free for a limited time on the DTMaaS platform
Figures
| Measure | Value | Measured by |
|---|---|---|
| Token consumption reduction vs LLMs | more than 90% customer service, risk control and financial scenarios | company |
| Response speed vs LLMs | up to 10x faster same tasks | company |
| Open-source benchmarks outperforming other decision models | 23 including Microsoft-Decision-1 and Jev 1.13; 27B reported highest accuracy at its size | company |
Results are the company's own and not independently verified. Access type is inferred from availability on the DTMaaS platform; weights are not mentioned.
Sources
This record was checked against its sources on 23 October 2026. How we check