Perplexity Releases Decider v1.3 Open-Weight Decision Models With Locate

Executive Briefing
“Perplexity has released the Decider v1.3 and Flash v1.3 open-weight decision models,...”
Key Takeaways
- 1.Perplexity released the open-weight decision models pplx-decider-v1.3 and pplx-decider-flash-v1.3 under the Apache 2.0 license on Hugging Face.
- 2.The new Locate feature extracts exact text spans by type or instruction by training a span head plus a LoRA adapter while keeping the backbone frozen.
- 3.The v1.3 model has 27B parameters based on Qwen3.8-27B, while Flash has 4B parameters distilled from the larger model.
Perplexity has released the Decider v1.3 and Flash v1.3 open-weight decision models, returning probabilities instead of generated text. The update introduces a Locate feature for extracting precise text spans.
Key Developments and Technical Context
Perplexity released the open-weight decision models pplx-decider-v1.3 and pplx-decider-flash-v1.3 under the Apache 2.0 license on Hugging Face.
The new Locate feature extracts exact text spans by type or instruction by training a span head plus a LoRA adapter while keeping the backbone frozen.
The v1.3 model has 27B parameters based on Qwen3.8-27B, while Flash has 4B parameters distilled from the larger model.
Strategic Analysis and Industry Implications
Converting a large language model architecture into a probability-returning classifier helps reduce hallucination and optimizes costs for automated query routing or classification tasks.
Outlook and Key Factors to Watch
All benchmark evaluation results are self-reported by Perplexity without independent third-party verification.







