动察 Beating AI News: Perplexity has launched the Decisions API and open-sourced the multimodal decision model pplx-decider-v1-27b. Instead of generating ordinary text, it directly returns a probability distribution over the answers provided by the developer, and can be used for classification, request routing, and selecting an Agent's next action. The API costs $0.04 per 1 million input tokens, and output is free.In the 11 evaluations released by Perplexity, totaling 7,210 samples, pplx-decider-v1-27b achieved an overall accuracy of 85.71%, higher than Jev's 84.51% and Qwen3.8-27B's 74.76%. However, it is not ahead across the board. Jev scored higher on 6 of the 11 tests, while Perplexity led on FinancialPhraseBank, RAGTruth, TabFact, and Circa. These results come from Perplexity's own testing, not an independent third-party evaluation.This type of model is specifically designed to handle the high-frequency "small judgments" within Agents. For example, which person a customer service ticket should be assigned to, whether a tool should be called, and which process to follow next. Traditional LLMs usually need to generate a passage of text and then parse the answer, whereas decision models directly return preset options and probabilities. AWS, Cloudflare, and others have also recently launched similar models.