Perplexity Ships Decider v1.3 Decision Models With Span Extraction as Decision-Model Category Crowds Up
The open-weight 27B and 4B models return probabilities instead of text and add Locate, an exact-span extraction feature, on 13 November.
Perplexity released Decider v1.3 (27B) and Flash v1.3 (4B) on 13 November, open-weight under Apache 2.0. Unlike generative models, Decider answers yes/no, multiple-choice and scoring questions by returning calibrated probabilities per option, and can also work over images. The v1.3 update adds Locate, which extracts exact text spans with character offsets for tasks such as named-entity recognition, PII detection and extractive QA. The company reports 91.6% mean accuracy across 63 classification tasks, though these figures are self-reported.
The release lands in a fast-crowding category: AWS Strands Decider 2B, Drex, Sarvam Prism and Perplexity have all shipped decision models within roughly six weeks. The pattern emerging across the industry separates a cheap, calibrated, auditable decision layer from an expensive generative layer - an architecture directly relevant for document pipelines that need per-field confidence and compliance-grade audit trails rather than fluent prose.