According to CoinMeta, Perplexity has released a new generation of decision-making models: Decider and v1.3, as well as a lightweight version called Decider Flash and v1.3. These models are based on Qwen3.8-27b and Qwen3.5-4b respectively. They are specifically designed for tasks such as classification and scoring, and do not need to generate answers word for word like traditional large models. The new versions include a feature called locate, which allows users to specify what they are looking for in natural language without having to write extraction rules in advance. A single request can simultaneously complete classification, scoring, and information extraction. Self-tests by Perplexity show that in 17 cross-domain information extraction tasks, Decider and v1.3 scored 69.0, surpassing the 43.3 of the control model Liquid D1. In 63 internal classification tasks, the average accuracy rate reached 91.6%, which is 6.54 percentage points higher than the company's current classification system.