Decider V1.2

Perplexity Decider V1.2 is a decision model that reads text and images and returns probabilities or scores for predefined questions, supporting tasks such as request routing and retrieval filtering. Compared with its predecessor, it uses a joint request format that processes related questions about the same state together. The developer says this lets related decisions share context, but recommends separate requests for independent retrieval candidates because joint scoring can make their results influence each other. A public model card describes weight-based deployment, although download access and a dedicated inference setup must be checked before adoption. The developer also reports overconfident probabilities on held-out data, so teams should calibrate outputs on their own cases and define when a decision needs human review.