Perplexity Ships Decider v1.3, Apache-Licensed Qwen-Based Classifier Hits 91.6% on 63 Internal Tasks

Perplexity released Decider v1.3 on Nov 12, a general-purpose classification model shipping in two sizes: Decider v1.3 (Qwen 27B) and Decider Flash v1.3 (distilled Qwen 4B). It hits 91.6% mean accuracy on 63 internal classification tasks, 6.54pp above production classifiers, and 88.16% across eight languages. Weights are Apache-licensed on Hugging Face and served via OpenRouter and Perplexity's own Decisions API at roughly $26/million requests locally.