Last updated: October 29, 2026
The Perplexity Decisions API gained two new decision models in October 2026: pplx-decider-v1.2, a 27B model at $0.02 per million input tokens, and pplx-decider-flash-v1.2, a new 4B model at $0.01. Both read text, JSON and images and return probabilities your code can act on, and output tokens are free.
Perplexity launched the Decisions API earlier in October and has updated its model twice, per the Perplexity API changelog. This guide explains what changed in version 1.2, how to call it, what it costs, and how to run the Flash model on a Mac.
Contents
- What’s new
- Models, pricing and limits
- How to make your first request
- Grouping questions and prompt patterns
- Run it locally with Decider for Mac
- How v1.2 compares with v1.1
- Limitations
- Frequently asked questions
What’s new in the Perplexity Decisions API
A decision model does not write replies, code or explanations. Per the Decisions API quickstart, you send content as state, attach named questions, and get back one typed answer per question:
noul: a yes or no question, answered with the probability of yes from 0 to 1.choice: one of up to 255 options you define, with a probability for every option.score: a level on an ordered rubric of up to 10 levels, with an expected score.
The October changes, in order:
- Launch: the API opened with
pplx-decider-v1-27bat $0.04 per million input tokens. - v1.1:
pplx-decider-v1.1-27breplaced it and cut input to $0.02. - v1.2:
pplx-decider-v1.2replaced v1.1 at the same price, and the newpplx-decider-flash-v1.2arrived at $0.01. Aliasespplx-decider-latestandpplx-decider-flash-latestfollow the newest version.
The pplx-decider-v1.2 model card on Hugging Face says the main change is a joint request format: the model reads the state and every question in one sequence, so several questions about the same content cost about as much as one. The training data was also reworked.
Models, pricing and limits
Both models take the same request and return the same response, so switching is a one-line change to model. Figures below come from Perplexity’s quickstart.
| Feature | pplx-decider-v1.2 | pplx-decider-flash-v1.2 |
|---|---|---|
| Alias | pplx-decider-latest | pplx-decider-flash-latest |
| Parameters | 27B | 4B |
| Input price | $0.02 per 1M tokens | $0.01 per 1M tokens |
| Output tokens | Free | Free |
| Input types | Text and images | Text and images |
| Context window | 256K tokens | 256K tokens |
| Several questions per request | Answered together; state billed once | Answered one by one; state billed per question |
| Tokens per minute | 15,000,000 | 4,000,000 |
Both models allow 1 to 128 questions per request, a 32 MiB body and up to 8 images per question. Each organization gets 10 requests per second by default.
Images must be PNG, JPEG or WebP data URLs; the API never fetches image URLs. Oversized images return a 504 after about a minute, so resize first.
How to make your first request
Any Perplexity API key works with the Decisions API. The quickstart steps are:
- Create or copy an API key in the Perplexity console and export it as
PERPLEXITY_API_KEY. - Install an SDK: the Python library 0.44.0 or later (
pip install "perplexityai>=0.44.0") or the TypeScript SDK 0.39.0 or later. Any HTTP client also works. - Send a POST to
https://api.perplexity.ai/v1/decisionswithmodel,stateandquestions, using a Bearer token in the Authorization header. - Read each answer under the name you gave its question, and log the
x-request-idheader for support. - Set thresholds on a labeled sample of your own data, leaving some margin because identical requests can differ slightly.
from perplexity import Perplexity
client = Perplexity(timeout=30.0)
r = client.decisions.create(
model="pplx-decider-latest",
state={"review": "The battery stopped charging after two weeks."},
questions={
"defect": {"type": "noul",
"instructions": "Does the review report a product defect?"},
"severity": {"type": "score",
"instructions": "How severe is the problem?",
"criteria": ["Cosmetic", "Inconvenient", "Product unusable"]},
},
)
print(r.answers["defect"].noul, r.answers["severity"].score)Note that confidence is calculated from the probabilities and is not the top probability. A key sent in x-api-key returns 401.
If you need written, cited answers instead of probabilities, Perplexity’s Agent API serves models such as the one in our Gemini 4 Argon power-user guide.
Grouping questions and prompt patterns
Grouping is the biggest practical change in v1.2. Perplexity’s guidance:
- Group facets of one decision. Several attributes of one item, “select all that apply” labels, or the steps of a routing decision belong in one request on pplx-decider-v1.2.
- Separate competing candidates. One question per document, tool or function call should go in separate requests, because grouping can lower their scores.
- When unsure, send one candidate per request. The model card says this is never more than about 0.4 points worse, while wrong grouping can cost about 3.
- Keep the grouping fixed when you tune thresholds.
- Use Flash with one question per request. Three questions about a 10,000-token state bill about 30,000 tokens on Flash, so with three or more questions the larger model usually costs less.
Run it locally with Decider for Mac
Decider for Mac is a menu bar app that runs pplx-decider-flash-v1.2 on your Mac and serves the same API at http://127.0.0.1:8080/v1/decisions. It needs no API key and has no usage bill, and your content stays on the machine.
Requirements are an Apple silicon Mac (M1 or later), macOS 13 or later, 16 GB of memory and about 6 GB of free disk space. The app downloads a 5.8 GB model on first launch and may ask for a Hugging Face read token.
Moving to the hosted API is a URL change plus an Authorization header. The local weights are a q8_0 GGUF build, so recheck locally tuned thresholds against the hosted model. Windows, Linux and Intel Macs are not supported.
For other open-weight releases this month, see our Mistral Large 4 guide and DeepSeek V4.1 Flash guide.
How v1.2 compares with v1.1
On Perplexity’s own public Decision Index 0.3, the model card reports an overall score of 68.09 for pplx-decider-v1.2 with joint requests, against 62.30 for pplx-decider-v1.1-27b. On the Vision Decision Index, the score rose from 73.22 to 77.43. These are vendor-reported results.
Version 1.1 answered one question per forward pass. Version 1.2 keeps the Qwen3.8-27B backbone but answers all questions in one sequence, so grouping now affects both cost and accuracy.
Compared with asking a general chat model for a label, the Decisions API avoids output parsing and returns numbers you can threshold. For general-purpose models that write text, see our Kimi K3.1 Pro guide.
Limitations
- It returns only probabilities, not explanations or text.
- Probabilities can be somewhat over-confident on unfamiliar tasks, according to the model card, so validate thresholds on your own data.
- Aliases can move to a newer model, and the response echoes the name you sent rather than the serving version.
- Large inputs are slow: Perplexity measured about 53 seconds for roughly 248,000 input tokens on pplx-decider-v1.2 and 16 seconds on Flash.
Frequently asked questions
What is the Perplexity Decisions API?
It is a Perplexity API endpoint, POST /v1/decisions, that answers questions about text, JSON or images with probabilities instead of generated text. You get a yes probability, one of your options, or a level on your rubric, which your code compares with a threshold.
How much does the Perplexity Decisions API cost?
Input costs $0.02 per million tokens on pplx-decider-v1.2 and $0.01 per million tokens on pplx-decider-flash-v1.2. Output tokens are free and there is no per-request fee, so the usage.input_tokens value in each response gives you the cost.
What is the difference between pplx-decider-v1.2 and the Flash model?
pplx-decider-v1.2 is the 27B default model and answers all questions in a request together, billing the state once. pplx-decider-flash-v1.2 is a 4B model at half the price per token, but it answers each question separately and bills the state once per question.
Can I run pplx-decider on my own computer?
Yes. Decider for Mac runs pplx-decider-flash-v1.2 locally on Apple silicon Macs with macOS 13 or later and 16 GB of memory, and serves the same API at 127.0.0.1 with no API key and no usage bill. The model cards on Hugging Face list an Apache 2.0 license.
Do old pplx-decider model names still work?
Yes. Requests that name pplx-decider-v1.1-27b or pplx-decider-v1-27b continue to work. Perplexity routes them to pplx-decider-v1.2 and bills them at its rate, so you should recheck any thresholds you tuned on the older model.
Sources
- Perplexity docs: Decisions API quickstart
- Perplexity docs: API changelog (October 2026 entries)
- Perplexity docs: Decider for Mac
- Hugging Face: perplexity-ai/pplx-decider-v1.2 model card
Image: Perplexity AI logo, public domain, via Wikimedia Commons. Placed on a white background and resized. Perplexity is a trademark of Perplexity AI, Inc.
