Perplexity Releases Decider v1.3 With Span-Extraction Feature
Perplexity has released pplx-decider-v1.3 and pplx-decider-flash-v1.3, open-weight decision models that return probabilities for labeled options rather than generated text. They support yes/no, multiple-choice and scoring tasks involving text or images. The weights are available under Apache 2.0 on Hugging Face, and both models can be accessed through the Perplexity Decisions API and OpenRouter.
Version 1.3 introduces Locate, which identifies exact text spans and returns their starting and ending character offsets. It is designed for named-entity recognition, personally identifiable information detection and extractive question answering. Users can specify a field label or an extraction instruction. The feature uses a separate span head and adapter while leaving the main decision components frozen.
Perplexity reports a mean accuracy of 91.6% across 63 query-classification tasks, along with 83.99 nDCG@10 on 72 retrieval and reranking tasks. Locate averaged 69.0 on 17 out-of-domain tasks. On one million real-world requests, the larger model agreed with frontier-model judges on 93.98% of questions. These results are self-reported. The 27B model requires about 49 GiB for BF16 weights, while the 4B Flash model requires about 8.5 GiB. API input pricing is $0.02 and $0.01 per 1M tokens, respectively.
Written by Sakhanda Wire AI from reporting by MarkTechPost. Facts, quotes and full context are in the original article below.