Perplexity's Decider V1.2 models now available on AI Gateway
By Steven Van ·
Two Perplexity decision models, a 27B and a 4B, are on AI Gateway with no markup, callable through the AI SDK, Decisions API or TypeSafe-compatible API.
Perplexity's Decider V1.2 27B and Decider Flash V1.2 4B are now available on Vercel's AI Gateway. Both are multimodal decision models for classifying content, routing requests, and scoring text, structured records, and images. Vercel documents a TypeSafe-compatible API alongside the AI SDK and the Decisions API for calling them.
How the models answer
The models answer typed questions with probabilities:
- Boolean questions return the likelihood of yes.
- Choice questions select from categories you define.
- Score questions assess content against an ordered rubric.
An application can use the results to route a support ticket, flag a damaged product, or send an uncertain decision for human review. Multiple questions can share the same input in one request.
Two model sizes
- Decider V1.2 27B (perplexity/pplx-decider-v1.2) is the larger model, for nuanced judgments and detailed rubrics.
- Decider Flash V1.2 4B (perplexity/pplx-decider-flash-v1.2) is the smaller model, for high-volume labeling, filtering, and routing with clear criteria.
Calling the models
The release note shows the same delivery-report check through the AI SDK, the Decisions API, and the TypeSafe-compatible API. In the AI SDK example, experimental_decide is imported from ai and takes a model, a state string, and a set of questions. For image input, base64 image_url objects are embedded in JSON state through the AI SDK or the TypeSafe-compatible API.
Existing TypeSafe clients can be pointed at AI Gateway by changing the API key and setting the base URL to https://ai-gateway.vercel.sh/typesafe. The compatible API supports POST /typesafe/v1/systemone and GET /typesafe/v1/models. A decision fallback option can rerun an uncertain decision with another model.
Billing and monitoring
Decision requests are recorded in request logs and included in spend budgets. Inference uses provider pricing with no markup or platform fee, including when you supply your own provider credentials through BYOK.