# What Is Perplexity Decider? The pplx-decider Models and Mac App, Explained

URL: https://whatisagenticai.net/blog/what-is-perplexity-decider/
Published: 2026-10-25
Publisher: Little Tough LLC, maker of AI Agent Assistant: Agentic AI (https://apps.apple.com/us/app/ai-agent-assistant-agentic-ai/id6761501260)

Perplexity Decider (pplx-decider) is an open decision model that answers with probabilities, not text. Now it runs free on your Mac. Versions, price, limits.

Perplexity Decider is the name people use for **pplx-decider**, a family of AI models from Perplexity that do one job: they read some text or an image and answer a closed question with a probability. They do not write paragraphs. Ask "Is this email a refund request?" and you get back a number like 0.93, not an essay.

The term is suddenly everywhere because Perplexity just shipped **Decider for Mac**, a free menu bar app that runs the small version of the model on your own computer, with no API key. Here is what Decider is, how the versions differ, what it costs, and how it compares to Jev and the OpenAI Decisions API.

**TL;DR:** Perplexity Decider (pplx-decider) is an open-weight "decision model" that returns yes/no probabilities, a pick from your options, or a score, instead of generated text. The first version, pplx-decider-v1-27b, came out on October 1, 2026, under the Apache 2.0 license; the current versions are pplx-decider-v1.2 (27B) and the small pplx-decider-flash-v1.2. Decider for Mac, announced October 23, runs the Flash model free on Apple silicon Macs with 16 GB of memory. It is a tool for developers and agent builders, not a chatbot.

## What is Perplexity Decider?

Most AI models are writers. You send a prompt, and they produce words one at a time. That is useful for drafts, but it is slow and messy when software only needs a label such as "spam" or "not spam."

A decision model skips the writing. Perplexity’s Decisions API takes two things: a state (the content to look at, such as a support ticket, a web page or a screenshot) and one or more named questions. Each question has one of three types:

| Question type | What you get back | Everyday example |
| --- | --- | --- |
| noul (yes/no) | A probability that the statement is true | Does this photo show a damaged package? |
| choice | One option from your list, with probabilities | Send this message to billing, tech support or sales |
| score | A level on a rubric you write | How urgent is this ticket, from "can wait" to "site is down"? |

The probabilities are the point. A program can act on its own when the model is 98% sure and pass the case to a person when it is 55% sure. Perplexity calls these calibrated probabilities, meaning a 0.8 should be right about 8 times out of 10.

## Perplexity Decider versions: v1, v1.1 and v1.2

Perplexity has moved fast. In about three weeks it shipped three versions, which is why people search for "perplexity decider v1.1" and "pplx-decider-v1-27b" side by side.

| Model | Size | What changed |
| --- | --- | --- |
| pplx-decider-v1-27b | 27B | First open release, October 1, 2026. Fine-tuned from Qwen3.8-27B, Apache 2.0, text and images. |
| pplx-decider-v1.1-27b | 27B | Early October update. Perplexity said it cost half as much as v1 and led Hugging Face’s Decision Index 0.3. |
| pplx-decider-v1.2 | 27B | Reads all questions in one pass. Model card score on Decision Index 0.3: 68.09, up from 62.30 for v1.1. |
| pplx-decider-flash-v1.2 | 4B (built on Qwen3.5-4B) | Small, fast version distilled from the 27B models. Decision Index 0.3: 58.88. This is the one Decider for Mac runs. |

All of these come with a context window of about 262,000 tokens, though Perplexity’s model cards note that training used inputs of up to about 32,000 tokens, and longer inputs have only been spot-checked. The benchmark numbers above are Perplexity’s own figures from its model cards.

## What is Decider for Mac?

Decider for Mac is a small app that sits in your Mac’s menu bar and runs pplx-decider-flash-v1.2 locally. Perplexity posted it on its developer forum on October 23, 2026. Once it is running, any program on your Mac can send requests to http://127.0.0.1:8080/v1/decisions, the same format as Perplexity’s hosted API.

What you need, according to Perplexity’s docs:

- A Mac with Apple silicon (M1 or later). Intel Macs, Windows and Linux are not supported.

- macOS 13 or later, 16 GB of memory and about 6 GB of free disk space.

- An internet connection on first launch, when the app downloads the 5.8 GB model.

The big selling point is privacy. The model reads your content on your own machine, so customer data, meeting notes or screenshots of internal tools never leave the computer. When you outgrow it, you change the URL to https://api.perplexity.ai/v1/decisions, add an API key, and get the larger 27B model. One caution from Perplexity: the Mac build is a compressed (q8_0) copy of the Flash model, so its answers can differ slightly from the hosted version. Check your thresholds before relying on it in production.

## Is Perplexity Decider free?

Mostly, yes. The weights are open under Apache 2.0, which allows commercial use, and you can run them yourself through Hugging Face or Ollama. Decider for Mac has no API key and no usage bill. The hosted API charges only for input: Perplexity listed v1.1 at $0.02 per million input tokens, half the price of v1, and reports put v1.2 at the same price with Flash at half that. As with other decision APIs, the answers themselves are not billed.

## Perplexity Decider vs Jev vs OpenAI Decisions API

Perplexity is not alone. A small race has formed around "System One" models that answer instantly instead of thinking out loud.

|  | Perplexity Decider | TypeSafe Jev | OpenAI Decisions API |
| --- | --- | --- | --- |
| Open weights | Yes (Apache 2.0) | No | No |
| Runs on your Mac | Yes, via Decider for Mac | No | No |
| Images | Yes | Not offered, per early comparisons | Yes |
| Hosted input price | $0.02 per 1M tokens (v1.1) | $0.042 per 1M tokens | $0.10 per 1M tokens |
| Launched | October 1, 2026 | September 15, 2026 | GA October 22, 2026 |

On its own 11-benchmark panel, Perplexity reported 85.7% for v1 against 84.5% for [Jev](https://whatisagenticai.net/blog/what-is-jev-ai/). That is a vendor test, so treat it as a claim, not a verdict. For the closed alternatives, see our explainers on the [OpenAI Decisions API](https://whatisagenticai.net/blog/what-is-openai-decisions-api/) and [Microsoft Decision-1](https://whatisagenticai.net/blog/what-is-microsoft-decision-1/).

## What this means for you

Decision models are the quick reflexes of AI agents. An agent makes hundreds of small calls while it works: Is this the right tool? Is this page safe? Should a person check this? A model that answers in a blink, with a confidence number, lets the agent check each step without slowing to a crawl. Decider for Mac pushes that idea onto ordinary laptops.

Unless you write code, you will not open Decider yourself. You will feel it through faster, more careful agents. If you want that today on your phone, **Agentic AI** gives you a team of expert agents on iPhone and iPad that plan the steps and deliver the result, with no account to create and 3 free tasks to start. New to the words? Try our [agentic AI glossary](https://whatisagenticai.net/agentic-ai-glossary/) or [what agentic AI is](https://whatisagenticai.net/what-is-agentic-ai/).

## Quick answers

**What is Perplexity Decider?** A family of open-weight Perplexity models, called pplx-decider, that answer closed questions about text or images with probabilities, a choice from your list, or a score, instead of writing text.

**What is pplx-decider-v1-27b?** The first Decider model, released October 1, 2026. It is a 27B model fine-tuned from Qwen3.8-27B and licensed under Apache 2.0. Newer versions are v1.1 and v1.2.

**What is new in Perplexity Decider v1.1 and v1.2?** v1.1 halved the hosted price to $0.02 per million input tokens. v1.2 reads all questions in one pass, scores higher on Decision Index 0.3, and added the small Flash model.

**Can I run Perplexity Decider locally?** Yes. Decider for Mac runs the Flash model on Apple silicon Macs with macOS 13 and 16 GB of memory. The 27B model is also on Ollama and Hugging Face for bigger machines.

**Is Perplexity Decider a chatbot?** No. It does not chat or write answers. It is a building block that developers use inside apps and AI agents.

**Sources:** [Perplexity docs: Decider for Mac](https://docs.perplexity.ai/docs/decisions/decider-for-mac) · [Perplexity forum announcement](https://community.perplexity.ai/t/you-can-now-run-pplx-decider-flash-v1-2-locally-on-your-mac-without-an-api-key/6413) · [pplx-decider-v1.2 model card](https://huggingface.co/perplexity-ai/pplx-decider-v1.2) · [pplx-decider-flash-v1.2 model card](https://huggingface.co/perplexity-ai/pplx-decider-flash-v1.2) · [Perplexity Developers on X (v1.1)](https://x.com/perplexitydevs/status/2107519531711418597) · [AlphaSignal](https://alphasignal.ai/news/perplexity-releases-pplx-decider-a-27b-open-model-that-replaces-label-parsing) · [Pasquale Pillitteri](https://pasqualepillitteri.it/en/news/26892/perplexity-decider-for-mac-local-decision-model) · [Ollama library](https://ollama.com/library/pplx-decider)