Perplexity puts two Decider models on AI Gateway, and boolean questions come back as probabilities
Synopsis
Perplexity's Decider V1.2 27B and Decider Flash V1.2 4B, two multimodal decision models, are now available through AI Gateway. They classify content, route requests and score text, structured records and images by answering typed questions with probabilities: boolean questions return the likelihood of yes, choice questions pick from categories the caller defines, and score questions assess content against an ordered rubric. An application can use those results to route a support ticket, flag a damaged product, or send an uncertain decision to human review, and several questions can share one input in a single request.
Interpretation
Decision models are now offered as typed questions that return probabilities rather than a single label, so an application can decide whether to act automatically or escalate. The prior pattern treated classification or scoring as a one-shot output; now boolean questions return the likelihood of yes, choice questions select among caller-defined categories, and score questions evaluate content against an ordered rubric. Perplexity's example feeds a delivery report as input, asks a boolean question about whether it reports damage, and reads that answer from the returned results.
The two models split the work by judgment difficulty and throughput, the larger for nuanced judgments and detailed rubrics, the smaller for high-volume labeling, filtering and routing with clear criteria. One family now offers two sizes, letting callers choose between nuanced judgment and bulk processing according to task complexity. Decider V1.2 27B covers nuanced judgments and detailed rubrics, while Decider Flash V1.2 4B covers high-volume labeling, filtering and routing; both accept text, structured records and images.
Calls go through AI Gateway, where decision requests are recorded in request logs and counted in spend budgets, and inference uses provider pricing without markup. Callers previously had to manage keys and costs separately; now one API key reaches the models and usage and cost are monitored in one place. The same pricing applies when callers supply their own provider credentials through BYOK; image input requires base64 image_url objects embedded in the JSON state.
Perspective
The report covers what Perplexity announced at availability: two models callable through AI Gateway, supporting boolean, choice and score questions, accepting text, structured records and images, with example calls and pricing notes. It is aimed at engineers working on content classification, request routing and human-review flows.
The source gives no accuracy, latency or comparison with other classification approaches, and does not say how the two models differ on specific tasks; where to set the probability threshold for human review is left to the caller.