Perplexity adds text extraction and downloadable weights to Decider models

Perplexity released Decider v1.3 and Decider Flash v1.3 on November 12, adding exact text extraction and downloadable Apache-2.0 weights, Unite.AI reported. Both models are also available through OpenRouter and Perplexity's Decisions API. Decider returns probabilities for classification and scoring tasks rather than generating prose. Its new Locate feature returns matching text and character positions for uses including personal-information detection and extractive question answering. Locate works only with text and allows up to 128 extraction questions per request. Mixing extraction and decision questions requires separate computation passes, adding latency. Self-hosting requires a CUDA GPU and memory beyond the weights; Flash has a smaller hardware footprint.