Perplexity said in a developer post on X from 6:24 p.m. Oct. 1, 2026 that it was introducing a new Decisions API, powered by pplx-decider-v1-27b, which the company describes as a multimodal decision model trained to output a probability distribution over a fixed set of answers instead of text.
Perplexity’s documentation says a decision model is “built to make fast, structured decisions that software can use directly.” The docs say developers can send text, JSON, or images as the state, attach named questions, and get one answer per question back. They also list three question types: “noul” for yes-or-no checks, “choice” for picking among user-defined options, and “score” for rating content against an ordered rubric.
The same docs say the API returns typed answers with probabilities rather than text, and that it costs $0.04 per million input tokens while output tokens are free.
What Perplexity says about pplx-decider-v1-27b
In a separate developer post on X from 6:24 p.m. Oct. 1, 2026, Perplexity said pplx-decider-v1-27b is fine-tuned from Qwen3.8-27B, has a 250,000-token context window, and linked its Hugging Face model page. The Hugging Face model card also presents pplx-decider-v1-27b as a decision model fine-tuned from Qwen3.8-27B.
Perplexity’s developer account linked to Hugging Face weights for pplx-decider-v1-27b. Separately, in a post from CEO Aravind Srinivas, he wrote that the company was open-sourcing a multimodal decision model and offering it through the new API at 4 cents per million input tokens with free output tokens. Srinivas’ post uses the name “pplx-decider-27b,” while Perplexity’s developer post and docs use pplx-decider-v1-27b, so the supplied evidence does not resolve whether those names refer to the same model.
On performance, the available evidence is Perplexity’s own materials. The company’s developer account said the model scores 85.71% across benchmarks, and the model card shows an “Overall” score of 85.71% across 11 benchmarks while noting that those results were measured through the Perplexity API.
The product pitch in the docs is straightforward: use one request to ask multiple structured questions about the same input and get back probabilities for each answer, rather than a block of prose to parse.
