Perplexity's Decider V1.2 models now available on Vercel AI Gateway

Perplexity Decider V1.2 27B and Decider Flash V1.2 4B are now available on AI Gateway. Both are multimodal models that classify content, route requests, and score text, structured records, and images. Vercel describes the 27B model, identified as perplexity/pplx-decider-v1.2, as the larger option for nuanced judgments and detailed rubrics.

The smaller Decider Flash V1.2 4B, identified as perplexity/pplx-decider-flash-v1.2, is aimed at high-volume labeling, filtering, and routing with clear criteria. The models return probabilities for Boolean questions, select from developer-defined categories for Choice questions, and assess content against an ordered rubric for Score questions. Multiple questions can share one input in a single request.

Vercel's example calls the model through the AI SDK's experimental_decide function with a boolean question about a damaged delivery. Requests made through AI Gateway are logged and count toward spend budgets, and inference uses provider pricing with no markup or platform fee, including when bring-your-own-key credentials are used.

Developers building agents can route support tickets, flag damaged goods, or send uncertain decisions to a person using a model call instead of custom classification code.

Because decision requests appear in AI Gateway request logs and spend budgets, teams can track their cost alongside other model usage.

For image inputs, the source says base64 image_url objects are embedded in JSON state, so check the decision models documentation for the exact request format before relying on it.