Perplexity releases open-weight Decider v1.3 and Flash classifiers
The AMW Read
The release adds specialized open-weight decision models to Perplexity's products, with potential use in agent workflows; production impact remains unproven.
Perplexity releases open-weight Decider v1.3 and Flash classifiers
Perplexity released pplx-decider-v1.3 and pplx-decider-flash-v1.3 on Hugging Face on November 13. The models use fine-tuned Qwen architectures and are available under the Apache 2.0 license. They evaluate text against structured criteria and produce probability distributions and verbatim text extractions.
The release adds specialized decision models to a market where companies also use generative models for classification and routing. Perplexity's approach gives builders access to model weights for tasks centered on selecting and extracting information. It also extends the company's use of open-weight models: in September, AI Market Watch reported that its Portable Computer agent ran locally on a Qwen model.
For builders, the practical question is whether these models fit a specific classification or routing task. Teams can evaluate the probability outputs and extracted spans against their own criteria before incorporating them into a workflow. The source does not provide performance results, so the release alone does not establish accuracy or suitability for production use.


