Many language-based applications require a probability distribution over a finite set of options—for example, classification or multiple-choice scoring—rather than a free-form response. The new Chinese Decision family of models is built specifically for this setting, operating in Chinese and producing probabilities in a single forward pass.

The abstract does not provide technical details on architecture or training data, but it positions the models as a dedicated solution for probabilistic decision-making. No comparisons to other models or benchmarks are mentioned, so the reported capability stands alone.

This work addresses a practical gap: while general language models can be prompted to give probabilities, dedicated models may offer more reliable and efficient outputs for decision-focused tasks. The single-pass design suggests an emphasis on speed and simplicity.