Energy‑based models (EBMs) assign a scalar energy to each joint configuration of observed and predicted variables. In decision tasks the model must pick one answer from a finite list supplied with each input; inference reduces to comparing the energies of the options, and training lowers the energy of the correct option relative to the others. Standard EBMs only learn relative compatibility: adding the same input‑dependent constant to all energies leaves the predictions, the Gibbs distribution, and typical training losses unchanged, so the model lacks a reference level to decide whether any option is acceptable. ZeroPoint makes the empty answer a dedicated state with energy fixed at zero for every input, providing an explicit reference for all listed options. The energies of the options are constrained to be non‑negative depths that measure compatibility relative to the empty answer. All options and the empty answer share a single distribution; an option at zero depth receives zero probability. Consequently the decision to select an option or to abstain is part of ordinary inference and is trained with the negative log‑likelihood loss, without a separate rejection rule. The energy function is a small multilingual encoder that reads the input together with all options and is invariant to their order. Ablation studies show that most of the gain comes from training with examples where no option fits. With the same encoder and data, the fixed zero separates inputs with and without an acceptable option slightly better than a learned rejection class and clearly better than a thresholded softmax, while leaving the choice among options unchanged.\n\nReview: By anchoring the empty answer at zero energy, ZeroPoint embeds the reject decision directly into the energy comparison mechanism, simplifying the architecture and improving the model’s ability to recognise unacceptable inputs. The approach demonstrates a practical, lightweight solution for multilingual decision making.