Perplexity’s Decision Model Runs Free on a Mac, No API Key

Perplexity has put its small decision model on the Mac. On 22 October 2026 its developer account said: “You can now run pplx-decider-flash-v1.2 locally on your Mac, without an API key.” Perplexity’s own page for the app, Decider for Mac, promises “No API key and no bill.” The same model costs $0.01 per million input tokens on Perplexity’s hosted API, a tenth of the $0.10 OpenAI lists for its Decisions API, by our arithmetic.

A decision model does not write. Perplexity’s documentation describes it as “a class of model built to make fast, structured decisions that software can use directly”: it reads text and images and returns typed answers with probabilities, such as yes or no, one of a set of options, or a score on a rubric.

Business Pill · ON-DEVICE AI

A short Business Engineer explainer on AI models that run on the device rather than in the cloud. It teaches the general idea only and says nothing about Perplexity’s app.

What Perplexity Released

• The model. Decider for Mac runs pplx-decider-flash-v1.2, which Perplexity’s page calls “Perplexity’s 4B decision model,” on the user’s own machine and serves the same Decisions API locally, at a local address on port 8080 by default.
• Private data stays put. Perplexity’s page says “Private content stays on your Mac,” so users can classify files they cannot send to a hosted API, such as customer exports, meeting notes and logs.
• Text and images. The page says users can send screenshots and photos as well as text.
• Hardware. Perplexity lists a Mac with Apple silicon (M1 or later), macOS 13 or later, 16 GB of memory and about 6 GB of free disk space. Intel Macs, Windows and Linux are not supported; Perplexity points those users to the hosted API.
• Same request, two places. Per the page, moving code from the Mac to the hosted API takes a URL change and an Authorization header.

What Was Posted on X

Perplexity’s developer account announced that pplx-decider-flash-v1.2 can now run locally on a Mac without an API key. The post describes Decider for Mac as an app that runs the 4B decision model on device: ask a question about text or an image and get a probability back, all on the user’s machine.

Decision Models We Have Covered, Side by Side

Prices are list prices per million input tokens as each company published them. Perplexity, OpenAI, Microsoft and Subquadratic say output tokens are free. Perplexity figures are from its own documentation; the rest are as stated in our earlier pieces.

Sources: Perplexity Decisions API pricing and quickstart pages and the Decider for Mac page, read 24 October 2026; our pieces on OpenAI’s Decisions API (7 and 22 October), Microsoft-Decision-1 (9 October), SubQ Snap (12 October) and Strands Decider 2B (1 October).

By Our Arithmetic

By our arithmetic, OpenAI’s $0.10 per million input tokens is ten times the $0.01 Perplexity lists for its 4B model and five times the $0.02 for its 27B model. Microsoft’s and TypeSafe’s $0.042 is 4.2 times the 4B rate; Subquadratic’s $0.05 is five times. These are list prices only. They do not account for accuracy, speed or volume discounts, which the table does not compare.

What the Mac Version Is, and Is Not

Perplexity’s page says the weights on the Mac are a q8_0 “GGUF build of the model, not the hosted deployment.” It tells developers who tune a threshold locally to check it “on a sample of your data against the hosted model before you rely on it in production.” Downloading the app and the model needs a network connection, the page says; the model is fetched from Hugging Face.

The page’s examples are practical: labelling a folder of private files, sorting screenshots, and a pre-commit check that blocks code containing secrets. For the hosted service, Perplexity’s pricing page lists no per-request fee and free output tokens on both models.

Connecting the Coverage

Several companies have launched decision models since 1 October, per our coverage. Our Strands Decider piece covered Amazon’s 1 October release, a 2B model built by deleting the text-writing head of a Qwen3.5-2B model, released open source under Apache 2.0. Our OpenAI pricing piece reported on 7 October that OpenAI’s Decisions API charges $0.10 per million input tokens with no output charges.

Our Microsoft Decision-1 piece covered Microsoft’s 9 October model at $0.042 per million input tokens, available in Microsoft Foundry. Our SubQ Snap piece covered Subquadratic’s 12 October model at $0.05 per million input tokens, which the company says handles up to 384K tokens of context. And our OpenAI general-availability piece reported on 22 October that OpenAI’s $0.10 list price is about 2.4 times the $0.042 TypeSafe lists for Jev.

What’s new: the same decision model is now offered both as a hosted API and as a free app on a user’s own Mac, from one company.

The Structural Read

As we read it, Perplexity is offering one model two ways: hosted for throughput and the larger 27B model, and local for content that “can’t leave your Mac,” in its page’s words. Its page says moving between the two takes a URL change and an Authorization header.

As we read it, the local version carries a stated caveat: the Mac weights are a quantised build, and Perplexity tells developers to check thresholds “on a sample of your data against the hosted model before you rely on it in production.”

As we read it, the comparison in our table is on list price only. Perplexity’s pages, in what we read, publish no accuracy comparison with the other decision models.

Three Implications

FOR DEVELOPERS Perplexity’s page says the Mac app serves the same request as the hosted Decisions API, on Apple-silicon Macs with 16 GB of memory.

FOR BUYERS OF DECISION APIS Listed input prices in our table run from $0.01 to $0.10 per million tokens, and Perplexity, OpenAI, Microsoft and Subquadratic say output is free.

FOR PRIVACY-BOUND TEAMS Perplexity’s page names customer exports, meeting notes and logs as content the local app can classify without sending it to a hosted API.

The Business Engineer Lens

This story maps onto the Business Engineer framework The State of The Inference Economy.

The framework states: “Inference is a revenue engine: continuous, distributed across millions of applications and agents, incurring marginal cost with every query.”

As we read it, a decision call is one of those queries. Perplexity now lists one at $0.01 per million input tokens hosted and offers the same model with no bill on a user’s Mac, per its pages. The framework was published in March 2026 and does not discuss this release.

What Is Not Established

Perplexity’s pages, in what we read, do not publish accuracy benchmarks comparing its decision models with OpenAI’s, Microsoft’s or the others in the table, so the table compares list prices, not quality. The Mac build is quantised and Perplexity itself says thresholds should be checked against the hosted model. We have not tested the app. We did not contact Perplexity.

The Bottom Line

Perplexity’s Decider for Mac runs its 4B decision model, pplx-decider-flash-v1.2, on Apple-silicon Macs with no API key and no bill, per its own page, while the hosted version lists at $0.01 per million input tokens. That is a tenth of OpenAI’s $0.10 Decisions API list price, by our arithmetic, in a category where Amazon, Microsoft and Subquadratic launched decision models this month, per our earlier pieces.