Perplexity’s two decision models charge $0.01 or $0.02 per million input tokens, with free output tokens for developers automating classification, routing or scoring without generated prose.
Perplexity’s API prices the 4B pplx-decider-flash-v1.2 at $0.01 and the 27B pplx-decider-v1.2 at $0.02 per million input tokens. Vercel also offers both through its AI Gateway.
Their input can be text or images, and outputs include a classification, choice or score. Perplexity lists content labeling, filtering and routing tasks to AI agents as uses.
Several questions can be run against the same input at once. Perplexity advises sending one question per request to the 4B model because grouping questions can hurt its accuracy.
In Perplexity’s Decision Index 0.3 results, the 27B model scored 67.75 with individual questions and 68.09 with grouped questions. Its predecessor scored 62.30 with individual questions. The company says official training splits were included for some evaluated tasks and were decontaminated against the test sets.
Perplexity’s model cards list both releases under the Apache 2.0 license; the repositories require authorized access.
